1,075 102 64MB
English Pages [2490] Year 2018
Bertram Müller Sebastian I. Wolf Editors-in-Chief Gert-Peter Brüggemann · Zhigang Deng Andrew S. McIntosh · Freeman Miller W. Scott Selbie Section Editors
Handbook of Human Motion
Handbook of Human Motion
Bertram Müller • Sebastian I. Wolf Editors-in-Chief
Gert-Peter Brüggemann • Zhigang Deng Andrew S. McIntosh • Freeman Miller W. Scott Selbie Section Editors
Handbook of Human Motion With 522 Figures and 68 Tables
Editors-in-chief Bertram Müller Motion and More Barcelona, Spain
Section Editors Gert-Peter Brüggemann Institute for Biomechanics und Orthopedics German Sport University Cologne Cologne, Germany
Sebastian I. Wolf Clinic for Orthopedics and Trauma Surgery Center for Orthopedics Trauma Surgery and Spinal Cord Injury Heidelberg University Hospital Heidelberg, Germany Zhigang Deng Department of Computer Science University of Houston Houston, TX, USA
Freeman Miller Andrew S. McIntosh Wilmington, Delaware, USA McIntosh Consultancy and Research Sydney, NSW, Australia Australian Collaboration for Research into Injury in Sport and its Prevention (ACRISP) Federation University Australia Ballarat, VIC, Australia Monash University Accident Research Centre Monash University Melbourne, VIC, Australia W. Scott Selbie HAS-Motion Inc. Kingston, ON, Canada C-Motion Inc. Germantown, MD, USA ISBN 978-3-319-14417-7 ISBN 978-3-319-14418-4 (eBook) ISBN 978-3-319-14419-1 (print and electronic bundle) https://doi.org/10.1007/978-3-319-14418-4 Library of Congress Control Number: 2017957051 # Springer International Publishing AG, part of Springer Nature 2018 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Printed on acid-free paper This Springer imprint is published by the registered company Springer International Publishing AG, part of Springer Nature. The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Foreword
The field of human locomotion biomechanics has developed drastically in the last 50 years. I remember when I first began in biomechanics in the early 1970’s how underdeveloped the field of human motion analysis and the related methodologies were. There were very few places in the world that were involved in this field of study, and we had to develop appropriate methodologies often from scratch. I remember a discussion with experts as whether or not the impact forces measured with force plates were artifacts or real. I also remember the high-speed film cameras and the complicated and time-consuming film analysis. Today, human movement analysis is a highly developed field with many facets, excellent high-technology equipment, highly sophisticated mathematical methods of data analysis, and thousands of experts working all over the world in many different fields of applications. Additionally, human movement analysis is instructed in almost all universities in many different faculties (e.g., kinesiology, engineering, medicine, injury prevention, etc.). The Handbook of Human Motion reflects these developments and provides an introduction to the subject, as well as a description of the current state-of-the-art technologies and a vision for the future of the field. The various sections provide a broad representation of the different fields of application of movement analysis and allow cross-fertilization between these fields. Furthermore, the Handbook of Human Motion provides a wealth of information from outstanding experts in the field that can be used to explore this innovative field of study. The study of human movement is exciting and will develop in the next few decades even further using wearable sensors and chips that provide real-time information to the athlete and patient. The principal work of the human movement analyst is to provide appropriate interpretation of the wealth of generated results. This handbook helps in contributing to this development. Calgary, Summer 2017
Benno Nigg
v
Preface
In the beginning, the science of human movement was represented by just a few outstanding personalities. Nowadays, this is no longer the case as the number of experts in the field has increased exponentially. With this increase, the field has diversified into many different scientific areas and specialities. Working within different fields but meeting up regularly at conferences related to human motion capture, we both realized that there may be a huge amount of knowledge that we were not even aware of. Specialized textbooks are available for all of the different disciplines, but to our knowledge there is, as yet, no compendium that serves to overarch them all. Therefore, in conjunction with five other specialists in different fields and a publishing house that shared our philosophy, we began compiling this handbook. With well over 100 chapters now realized, and with more than 200 authors involved, the handbook has advanced to this first printed edition. This handbook is intended not only for beginners in the field, providing an overview of the different approaches, but also for experts in a particular area needing information for a related one. Some fields have developed along with the technology, such as animations in the motion picture industry. Others, such as medicine, are introducing it to improve understanding of human movement for diagnosis and treatment. Its application in the field of sports science, for instance, might enhance human performance while diminishing the risk of injury. When legal aspects are involved or when dealing with occupational medicine, the focus of human motion might shift toward other related issues and the subject of forensics. New fields are continually emerging, guided by the creativity of people working with such systems. In each chapter of this book, information is provided on state-of-the-art science, as well as insight into future directions, providing detailed information about each subject. Our hope is that this handbook will not only increase the level of knowledge of the individual reader, but also facilitate understanding between different scientific fields and related areas. The inclusion of authors from around the world not only reflects different approaches but also a wide variation in methodological and linguistic styles. The latter, for instance, is responsible for the transmission of information, which is the basis for communication and understanding in a multicultural society, and can involve imparting many ambiguous terms and concepts. While standardization
vii
viii
Preface
might not be an attainable goal, knowledge of the differences may well enhance interaction and communication between professionals. And finally, the variety of subject matter within this compact handbook might also produce new ideas for one’s own field of study. Interpolating knowledge with information from fields not previously believed to be related might just lead to novel approaches. What it does already provide is evidence of diversity in measurements and their interpretation, which will doubtless lead to improvements in results in the quest for objective data. What we, the editors, have also noted is the rapid development and expansion in the discovery of human motion and its application. While this first physical edition provides extensive information, it is also recommended to look up the online version, where frequent updates can be seen. We are thankful to all of the authors for their contribution to this edition and to their continuing interest in keeping the information up to date. We would also like to invite all of those who find any omissions to contribute to this work in the future. Barcelona/Heidelberg October 2017
Bertram Müller Sebastian I. Wolf
Acknowledgments
Editing this book was for both editors an adventure. We did not know in advance what result to expect since neither of our scientific backgrounds and networks overarched the disciplines presented in this book. We therefore give thanks to our section editors Scott Selbie, Freeman Miller, Gert-Peter Brüggemann, Zhigang Deng, and Andrew S. McIntosh for extending our network and helping in collecting these chapters. Our special thanks go to Manfred Nusseck and Benita Kuni for making the sections of human movement in music and dance possible. Furthermore, we wish to thank all contributing authors for their idealism in sharing their knowledge and dedicating their time to writing a contribution to this book. Finally, we would also like to extend our particular gratitude to the team of Springer with Tom, Lydia, Johanna, and Barbara for supporting us in realizing this overview on human motion throughout the past years.
ix
Contents
Volume 1 Section I Part I
Methods and Models
Rigid Body Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries . . . . . . . . . . . . . . . . . . . . . . Aurelio Cappozzo
3
Three-Dimensional Reconstruction of the Human Skeleton in Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valentina Camomilla, Aurelio Cappozzo, and Giuseppe Vannozzi
17
Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis . . . . . . . . . . . . . Raphaël Dumas and Janis Wojtusch
47
Part II
.....
79
3D Dynamic Pose Estimation from Marker-Based Optical Data . . . . . . W. Scott Selbie and Marcus J. Brown
81
Discriminative Methods in Dynamic Pose Estimation
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hans Gray, Shanyuanye Guan, Peter Loan, and Marcus Pandy 3D Musculoskeletal Kinematics Using Dynamic MRI . . . . . . . . . . . . . . Frances T. Sheehan and Richard M. Smith
101 117
Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Richard M. Smith and Frances T. Sheehan
135
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and Ligament . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Glen Lichtwark
157 xi
xii
Part III
Contents
Generative Methods in Dynamic Pose Estimation . . . . . . .
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thomas M. Kepple and Alan R. De Asha 3D Dynamic Pose Estimation from Markerless Optical Data . . . . . . . . Steven Cadavid and W. Scott Selbie Three-Dimensional Human Kinematic Estimation Using MagnetoInertial Measurement Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Andrea Cereatti, Ugo Della Croce, and Angelo M. Sabatini
177
179 197
221
Gait Parameters Estimated Using Inertial Measurement Units . . . . . . . Ugo Della Croce, Andrea Cereatti, and Martina Mancini
245
Physics-Based Models for Human Gait Analysis . . . . . . . . . . . . . . . . . . Petrissa Zell, Bastian Wandt, and Bodo Rosenhahn
267
Part IV
293
Body Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Scaling and Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . At L. Hof
295
Part V
307
Extended Modeling Techniques . . . . . . . . . . . . . . . . . . . . . . .......
309
Optimal Control Modeling of Human Movement . . . . . . . . . . . . . . . . . Brian R. Umberger and Ross H. Miller
327
Time Series Analysis in Biomechanics . . . . . . . . . . . . . . . . . . . . . . . . . . W. Brent Edwards, Timothy R. Derrick, and Joseph Hamill
349
Hill-Based Muscle Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ross H. Miller
373
Induced Acceleration and Power Analyses of Human Motion Anne K. Silverman
Simulation of Soft Tissue Loading from Observed Movement Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scott C. E. Brandon, Colin R. Smith, and Darryl G. Thelen
395
...
429
Gait Symmetry Measures and Their Relevance to Gait Retraining Silvia Cabral Part VI
Dynamic Electromyography . . . . . . . . . . . . . . . . . . . . . . . .
Surface Electromyography to Study Muscle Coordination . . . . . . . . . . François Hug and Kylie Tucker
449 451
Contents
xiii
Section II Part VII
Medical Application
Gait Assessment in Clinical Context
.................
471
Clinical Gait Assessment by Video Observation and 2D Techniques . . . Andreas Kranzl
473
The Conventional Gait Model - Success and Limitations . . . . . . . . . . . Richard Baker, Fabien Leboeuf, Julie Reay, and Morgan Sangeux
489
Variations of Marker Sets and Models for Standard Gait Analysis Felix Stief
...
509
Next-Generation Models Using Optimized Joint Center Location Ayman Assi, Wafa Skalli, and Ismat Ghanem
....
527
Kinematic Foot Models for Instrumented Gait Analysis . . . . . . . . . . . . Alberto Leardini and Paolo Caravaggi
547
Trunk and Spine Models for Instrumented Gait Analysis . . . . . . . . . . . Robert Needham, Aoife Healy, and Nachiappan Chockalingam
571
..........
583
Upper Extremity Models for Clinical Movement Analysis Andrea Giovanni Cutti, Ilaria Parel, and Andrea Kotanxis Part VIII
Interpreting Kinetics and EMG in Gait . . . . . . . . . . . . . . .
607
Interpreting Ground Reaction Forces in Gait . . . . . . . . . . . . . . . . . . . . Nachiappan Chockalingam, Aoife Healy, and Robert Needham
609
Interpreting Joint Moments and Powers in Gait . . . . . . . . . . . . . . . . . . L. H. Sloot and M. M. van der Krogt
625
EMG Activity in Gait: The Influence of Motor Disorders . . . . . . . . . . . Dimitrios A. Patikas
645
Part IX
671
Scores and Spatiotemporal Parameters . . . . . . . . . . . . . . .
Gait Scores: Interpretations and Limitations . . . . . . . . . . . . . . . . . . . . . Veronica Cimolin and Manuela Galli
673
Interpreting Spatiotemporal Parameters, Symmetry, and Variability in Clinical Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arnaud Gouelle and Fabrice Mégrot
689
Part X
709
Pedobarography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Assessing Pediatric Foot Deformities by Pedobarography . . . . . . . . . . . Dieter Rosenbaum
711
xiv
Contents
Assessing Clubfoot and Cerebral Palsy by Pedobarography . . . . . . . . . Julie A. Stebbins
727
Low Density Pedoboragraphy as a Gait Analysis Tool . . . . . . . . . . . . . Ruopeng Sun, Tyler A. Wood, and Jacob J. Sosnoff
741
The Importance of Foot Pressure in Diabetes . . . . . . . . . . . . . . . . . . . . Malindu E. Fernando, Robert G. Crowther, and Scott Wearing
759
Integration of Foot Pressure and Foot Kinematics Measurements for Medical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Claudia Giacomozzi, Paolo Caravaggi, Julie A. Stebbins, and Alberto Leardini
789
Volume 2 Part XI
Energy Consumption During Gait . . . . . . . . . . . . . . . . . . . .
811
Assessing the Impact of Aerobic Fitness on Gait . . . . . . . . . . . . . . . . . . Annet Dallmeijer, Astrid Balemans, and Eline Bolster
813
Oxygen Consumption in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . Hank White, J. J. Wallace, and Sam Augsburger
825
The Use of Kinematics for Pulmonary Volume Assessment . . . . . . . . . . Carlo Massaroni
847
Aerobic Capacity and Aerobic Load of Activities of Daily Living After Stroke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I. J. Blokland, T. IJmker, and H. Houdijk
863
Part XII
885
Gait and Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Measures to Determine Dynamic Balance . . . . . . . . . . . . . . . . . . . . . . . Timothy A. Niiler
887
Slip and Fall Risk Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Feng Yang
915
Detecting and Measuring Ataxia in Gait . . . . . . . . . . . . . . . . . . . . . . . . Mariano Serrao and Carmela Conte
937
Gait During Real-World Challenges: Gait Initiation, Gait Termination, Acceleration, Deceleration, Turning, Slopes, and Stairs . . . . . . . . . . . . Michael Orendurff Gait Retraining for Balance Improvement . . . . . . . . . . . . . . . . . . . . . . . Robert G. Crowther and Jessica May Pohlmann
955 977
Contents
Part XIII
xv
Pathoanatomy and Diagnostics in Cerebral Palsy . . . . . .
Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Freeman Miller and John Henley
987
989
Walking and Physical Activity Monitoring in Children with Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005 Kristie F. Bjornson and Nancy Lennon Spasticity Effect in Cerebral Palsy Gait . . . . . . . . . . . . . . . . . . . . . . . . . 1037 Marlene Cristina Neves Rosa and André Gonçalo Gomes Roque Natural History of Cerebral Palsy and Outcome Assessment . . . . . . . . 1053 Erich Rutz and Pam Thomason Skeletal Muscle Structure in Spastic Cerebral Palsy . . . . . . . . . . . . . . . 1075 Adam Shortland Part XIV
Movement Deviations in Cerebral Palsy
.............
1091
Swing Phase Problems in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . 1093 Ana Presedo Strength Related Stance Phase Problems in Cerebral Palsy . . . . . . . . . 1109 Justin Connor and Mutlu Cobanoglu Foot and Ankle Motion in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . 1121 Jon R. Davids and Sean A. Tabaie The Arm Pendulum in Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1151 Jaques Riad Upper Extremity Movement Pathology in Functional Tasks . . . . . . . . . 1167 Lisa Mailleux, Cristina Simon-Martinez, Hilde Feys, and Ellen Jaspers Part XV
Other Neurologic Gait Disorders . . . . . . . . . . . . . . . . . . . .
1185
Idiopathic Toe Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1187 Karen Davies, Lise Leveille, and Christine Alvarez Gait Disorders in Persons After Stroke . . . . . . . . . . . . . . . . . . . . . . . . . 1205 Johanna Jonsdottir and Maurizio Ferrarin Hereditary Motor Sensory Neuropathy: Understanding Function Using Motion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1217 Sylvia Õunpuu and Kristan Pierz Motor Patterns Recognition in Parkinson’s Disease . . . . . . . . . . . . . . . 1237 Pierpaolo Sorrentino, Valeria Agosti, and Giuseppe Sorrentino
xvi
Contents
Gait and Multiple Sclerosis James McLoughlin
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1253
Functional Dystonias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1267 Jessica Pruente and Deborah Gaebler-Spira Part XVI
Traumatic and Orthopedic Gait Disorders . . . . . . . . . . . .
1281
Gait Changes in Skeletal Dysplasia . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1283 William G. Mackenzie and Oussama Abousamra Impact of Scoliosis on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1289 Elizabeth A. Rapp and Peter G. Gabos Concussion Assessment During Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . 1307 Robert D. Catena and Kasee J. Hildenbrand Functional Effects of Ankle Sprain Ilona M. Punt and Lara Allet Part XVII
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 1325
Orthotics and Prosthetics in Gait . . . . . . . . . . . . . . . . . . .
1341
Prosthetic Foot Principles and Their Influence on Gait . . . . . . . . . . . . . 1343 Andrew Hansen and Felix Starker The Influence of Prosthetic Knee Joints on Gait . . . . . . . . . . . . . . . . . . 1359 Steven A. Gard Influence of Prosthetic Socket Design and Fitting on Gait . . . . . . . . . . . 1383 Arezoo Eshraghi and Jan Andrysek Functional Effects of Foot Orthoses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1407 Christopher Nester Functional Effects of Shoes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1423 Benedicte Vanwanseele Gait Rehabilitation with Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . . . 1433 Stefano Federici, Fabio Meloni, and Marco Bracalenti Brain-Computer Interfaces for Motor Rehabilitation . . . . . . . . . . . . . . 1471 Rüdiger Rupp Part XVIII
Gait After Joint Replacement . . . . . . . . . . . . . . . . . . . . .
1503
Effects of Total Hip Arthroplasty on Gait . . . . . . . . . . . . . . . . . . . . . . . 1505 Swati Chopra and Kenton R. Kaufman
Contents
xvii
Effects of Knee Osteoarthritis and Joint Replacement Surgery on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1521 Cheryl L. Hubley-Kozey and Janie Astephen Wilson The Effects of Ankle Joint Replacement on Gait . . . . . . . . . . . . . . . . . . 1551 Justin Michael Kane, Scott Coleman, and James White Brodsky Shoulder Joint Replacement and Upper Extremity Activities of Daily Living . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1563 Hendrik Bruttel, David M. Spranz, Jan M. Eckerle, and Michael W. Maier
Section III Part XIX
Arts and Human Performance
Sports
.........................................
Sprint Running: Running at Maximum Speed Michiyoshi Ae
1581
. . . . . . . . . . . . . . . . . . . 1583
Running Shoes: Injury Protection and Performance Enhancement . . . 1613 Steffen Willwacher Landings: Implications for Performance . . . . . . . . . . . . . . . . . . . . . . . . 1629 Laura A. Held, Henryk Flashner, and Jill L. McNitt-Gray Airborne Movements: Somersaults and Twists . . . . . . . . . . . . . . . . . . . 1661 Maurice R. Yeadon Ski Jumping: Aerodynamics and Kinematics of Take-Off and Flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1681 Mikko Virmavirta The Segmental Movements in Front Crawl Swimming . . . . . . . . . . . . . 1703 Ross H. Sanders, Jordan T. Andersen, and Hideki Takagi Movement Analysis of Scull and Oar Rowing . . . . . . . . . . . . . . . . . . . . 1719 Patria A. Hume Segmental Movements in Cycling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1741 Rodrigo R. Bini and Felipe P. Carpes Movement Analysis of the Golf Swing . . . . . . . . . . . . . . . . . . . . . . . . . . 1755 Patria A. Hume and J. Keogh The Motor Solutions of Throws in Sports . . . . . . . . . . . . . . . . . . . . . . . 1773 Bing Yu
xviii
Contents
Volume 3 Part XX
Motion in Music . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1787
Body Movements in Music Performances: The Example of Clarinet Players . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1789 Manfred Nusseck, Marcelo M. Wanderley, and Claudia Spahn Investigating Aspects of Movement in Violin Performance . . . . . . . . . . 1803 Gongbing Shan, Peter Visentin, Manfred Nusseck, and Claudia Spahn Movement and Touch in Piano Performance . . . . . . . . . . . . . . . . . . . . . 1821 Werner Goebl Movements, Timing, and Precision of Drummers . . . . . . . . . . . . . . . . . 1839 Sofia Dahl Observing and Learning Complex Actions: On the Example of Guitar Playing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1859 Tom Gardner and Emily S. Cross Part XXI
Dance
.........................................
1873
Functional Movement Analysis in Dance . . . . . . . . . . . . . . . . . . . . . . . . 1875 Andrea Schärli Motion Analysis as Pedagogic Tool in Dance . . . . . . . . . . . . . . . . . . . . . 1889 Martin Puttke and Dimitri Volchenkov Somatic Practices: How Motion Analysis and Mind Images Work Hand in Hand in Dance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1911 Sarah Whatley
Section IV Part XXII
Animation
Human Animation Control . . . . . . . . . . . . . . . . . . . . . . . .
1927
Video-Based Performance Driven Facial Animation . . . . . . . . . . . . . . . 1929 Fuhao Shi Depth Sensor-Based Facial and Body Animation Control . . . . . . . . . . . 1943 Yijun Shen, Jingtian Zhang, Longzhi Yang, and Hubert P. H. Shum Real-Time Full-Body Pose Synthesis and Editing Edmond S. L. Ho and Pong C. Yuen Real-Time Full Body Motion Control John Collomosse and Adrian Hilton
. . . . . . . . . . . . . . . . . 1959
. . . . . . . . . . . . . . . . . . . . . . . . . . 1975
Contents
Part XXIII
xix
Human Animation Generation . . . . . . . . . . . . . . . . . . . .
2001
Data-Driven Character Animation Synthesis . . . . . . . . . . . . . . . . . . . . . 2003 Taku Komura, Ikhsanul Habibie, Jonathan Schwarz, and Daniel Holden Physically Based Character Animation Synthesis . . . . . . . . . . . . . . . . . 2033 Jie Tan Biped Controller for Character Animation . . . . . . . . . . . . . . . . . . . . . . 2055 KangKang Yin, Stelian Coros, and Michiel van de Panne Data-Driven Hand Animation Synthesis Sophie Jörg
. . . . . . . . . . . . . . . . . . . . . . . . 2079
Example-Based Skinning Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . 2093 Tomohiko Mukai Part XXIV
Facial Animation and Gestures . . . . . . . . . . . . . . . . . . . .
2113
Visual Speech Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2115 Lei Xie, Lijuan Wang, and Shan Yang Blendshape Facial Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2145 Ken Anjyo Eye Animation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2157 Andrew T. Duchowski and Sophie Jörg Head Motion Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2177 Najmeh Sadoughi and Carlos Busso Hand Gesture Synthesis for Conversational Characters . . . . . . . . . . . . 2201 Michael Neff Laughter Animation Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2213 Yu Ding, Thierry Artières, and Catherine Pelachaud Part XXV
Crowd Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2231
Functional Crowds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2233 Jan M. Allbeck Crowd Formation Generation and Control . . . . . . . . . . . . . . . . . . . . . . 2243 Jiaping Ren, Xiaogang Jin, and Zhigang Deng Crowd Evacuation Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2255 Tomoichi Takahashi
xx
Part XXVI
Contents
Facial Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2269
Perceptual Study on Facial Expressions . . . . . . . . . . . . . . . . . . . . . . . . . 2271 Eva G. Krumhuber and Lina Skora Part XXVII
Human to Virtual-Human Interaction . . . . . . . . . . . . . .
2287
Utilizing Unsupervised Crowdsourcing to Develop a Machine Learning Model for Virtual Human Animation Prediction . . . . . . . . . . 2289 Michael Borish and Benjamin Lok
Section V
Forensics and Legal Application
Part XXVIII Functional Capacity Evaluation
..................
2307
Functional Capacity Evaluation and Preemployment Screening . . . . . . 2309 Elizabeth Chapman, Anne M. Felts, and Matthew Klinker Functional Capacity Evaluation and Quantitative Gait Analysis: Lower Limb Disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2331 Isabella T. Klöpfer-Krämer and Peter Augat Preparticipation Physical Evaluation in Sport . . . . . . . . . . . . . . . . . . . . 2349 James A. Onate and Daniel R. Clifton Part XXIX
Forensics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2361
Injury Mechanisms in Traffic Accidents . . . . . . . . . . . . . . . . . . . . . . . . 2363 Brian D. Goodwin, Sajal Chirvi, and Frank A. Pintar Vehicle Occupants in Traffic Accidents Garrett A. Mattos
. . . . . . . . . . . . . . . . . . . . . . . . . 2399
Slips, Trips, and Falls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2417 Andrew Short and Len Cubitt Biomechanical Forensics in Pediatric Head Trauma . . . . . . . . . . . . . . . 2447 Brittany Coats and Susan Margulies Head Impact Biomechanics of “King Hit” Assaults . . . . . . . . . . . . . . . . 2463 Declan A. Patton and Andrew S. McIntosh Expert Opinion and Legal Considerations . . . . . . . . . . . . . . . . . . . . . . . 2475 Henry M. Silvester Applications in Forensic Biomechanics Andrew S. McIntosh
. . . . . . . . . . . . . . . . . . . . . . . . . 2495
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2509
About the Editors
Bertram Müller started his professional life with two degrees in precision mechanics and electronics, he worked in the Institute of Experimental Physics at the Martin Luther University in Halle/Saale. A third engineering degree and an exchange program in a center for children with special needs in Australia awoke his interest in bridging Engineering and Medicine. Consequently, he undertook a Ph.D. in Bioengineering at Strathclyde University in Glasgow and graduated in Medical Science at University Rey Juan Carlos in Madrid. In 2000, he moved to Barcelona, where he managed the Biomechanics Laboratory of Egarsat until 2012. He presently divides his professional life between teaching and applied biomechanics. He is Associate Professor of Physics and Biomechanics at the University of Girona (EUSES), the University of Manresa, and the National Institute of Physical Education in Barcelona. He also works as a Consultant for applied biomechanics, including clinical and occupational biomechanics, as well as sports applications, having collaborated with the High-Performance Centre (CAR) in St. Cugat. His current research interest is functional capacity evaluation in various disciplines, including legal matters. He is Editor-in-Chief of the Journal of Forensic Biomechanics, as well as reviewer for several other scientific journals. Sebastian I. Wolf with a Ph.D. in Physics, Sebastian Wolf spent several years in pure physics research before he moved into the field of human movement analysis in 2001. For 10 years, he was Technical Director of the Gait Analysis Laboratory in the Department of Orthopedic Surgery at Heidelberg University. Since 2010, he has been the Director of the Division of Human Movement Analysis and is responsible for both the clinical gait analysis service as well as overseeing clinical research in this field. In 2015, he became Associate
xxi
xxii
About the Editors
Professor for Orthopedic Biomechanics at the Medical Faculty of Heidelberg University. In the same year, he was elected President of the European Society for Movement Analysis in Adults and Children (ESMAC). Sebastian Wolf has published more than 80 peer-reviewed articles relating to clinical movement analysis and is a reviewer for numerous journals in this field with continuing scientific interest in advancing knowledge on neurologic and orthopedic gait disorders and mobility-related medical healthcare.
Section Editors
Gert-Peter Brüggemann Institute for Biomechanics und Orthopedics, German Sport University Cologne, Cologne, Germany Zhigang Deng Department of Computer Science, University of Houston, Houston, TX, USA Andrew S. McIntosh McIntosh Consultancy and Research, Sydney, NSW, Australia Australian Collaboration for Research into Injury in Sport and its Prevention (ACRISP), Federation University Australia, Ballarat, VIC, Australia Monash University Accident Research Centre, Monash University, Melbourne, VIC, Australia Freeman Miller Wilmington, Delaware, USA Sebastian I. Wolf Clinic for Orthopedics and Trauma Surgery, Center for Orthopedics, Trauma Surgery and Spinal Cord Injury, Heidelberg University Hospital, Heidelberg, Germany W. Scott Selbie HAS-Motion Inc., Kingston, ON, Canada C-Motion, Inc., Germantown, MD, USA
xxiii
Advisory Panel
Benita Kuni Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany Benno Nigg Faculties of Kinesiology, Engineering and Medicine, University of Calgary, Calgary, AB, Canada Manfred Nusseck University of Music and University Clinic Freiburg, Freiburg, Germany Mark Stringer Clarivate, Barcelona, Spain
xxv
Contributors
Oussama Abousamra Nemours Alfred I. duPont Hospital for Children, Wilmington, DE, USA Michiyoshi Ae Faculty of Sport Science, Nippon Sport Science University, Tokyo, Japan Valeria Agosti Department of Motor Sciences and Wellness, University of Naples Parthenope, Naples, Italy Institute Hermitage-Capodimonte, Naples, Italy Jan M. Allbeck George Mason University, Fairfax, VA, USA Lara Allet Department of Physical Therapy, University of Applied Sciences of Western Switzerland, Carouge, Switzerland Department of Community Medicine, Geneva University Hospitals and University of Geneva, Geneva, Switzerland Christine Alvarez Shriners Gait Lab, Sunny Hill Health Centre for Children, Vancouver, BC, Canada British Columbia Children’s Hospital, Vancouver, BC, Canada Jordan T. Andersen Exercise and Sport Science, Faculty of Health Sciences, The University of Sydney, Sydney, NSW, Australia Jan Andrysek Holland Bloorview Kids Rehabilitation Hospital, Bloorview Research Institute, Toronto, ON, Canada Institute of Biomaterial and Biomedical Engineering, University of Toronto, Toronto, ON, Canada Ken Anjyo OLM Digital, Setagaya, Tokyo, Japan Thierry Artières Ecole Centrale Marseille, Marseille, France Laboratoire d’Informatique Fondamentale (LIF), UMR CNRS 7279, Université AixMarseille, Paris, France Ayman Assi Laboratory of Biomechanics and Medical Imaging, Faculty of Medicine, University of Saint-Joseph, Mar Mikhael, Beirut, Lebanon xxvii
xxviii
Contributors
Institut de Biomécanique Humaine Georges Charpak, Arts et Métiers ParisTech, Paris, France Janie Astephen Wilson School of Biomedical Engineering, Faculty of Medicine, Dalhousie University, Halifax, NS, Canada Nova Scotia Health Authority, Halifax, NS, Canada Peter Augat Berufsgenossenschaftliche Unfallklinik Murnau und Paracelsus Medizinische Privatuniversität Salzburg, Institut für Biomechanik, Murnau am Staffelsee, Germany Sam Augsburger Motion Analysis Center, Shriners Hospitals for Children Medical Center, Lexington, KY, USA Richard Baker University of Salford, Salford, UK Astrid Balemans Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, EMGO+ Institute for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands Brain Center Rudolf Magnus and Center of Excellence for Rehabilitation Medicine University Medical Center, Utrecht, The Netherlands De Hoogstraat Rehabilitation, Utrecht, The Netherlands Rodrigo R. Bini La Trobe Rural Health School, College of Science, Health and Engineering, La Trobe University, Bendigo, VIC, Australia Kristie F. Bjornson Seattle Children’s Research Institute, University of Washington, Seattle, WA, USA I. J. Blokland Faculty of Behavioral and Movement Sciences, Vrije Universiteit Amsterdam, Amsterdam Movement Sciences, Amsterdam, The Netherlands Heliomare Rehabilitation, Research and Development, Wijk aan Zee, The Netherlands Eline Bolster Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, EMGO+ Institute for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands Michael Borish Computer and Information Sciences and Engineering Department, University of Florida, Gainesville, FL, USA Marco Bracalenti Department of Philosophy, Social and Human Sciences and Education, University of Perugia, Perugia, Italy Scott C. E. Brandon Department of Mechanical Engineering, University of Wisconsin-Madison, Madison, WI, USA
Contributors
xxix
James White Brodsky Faculty, Foot and Ankle Fellowship Program, Baylor University Medical Center, Dallas, TX, USA University of Texas Southwestern Medical School, Dallas, TX, USA Texas A&M HSC College of Medicine, Bryan, TX, USA Marcus J. Brown HAS-Motion, Inc., Kingston, ON, Canada Hendrik Bruttel Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany Carlos Busso Multimodal Signal Processing Lab, University of Texas at Dallas, Dallas, TX, USA Silvia Cabral Laboratório de Biomec^anica e Morfologia Funcional, CIPER, Faculdade de Motricidade Humana, Universidade de Lisboa, Cruz Quebrada, Dafundo, Portugal Steven Cadavid KinaTrax Inc., Palm Beach, FL, USA Valentina Camomilla Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome Foro Italico, Rome, Italy Aurelio Cappozzo Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome “Foro Italico”, Rome, Italy Paolo Caravaggi Movement Analysis Laboratory and Functional-Clinical Evaluation of Prostheses, Istituto Ortopedico Rizzoli, Bologna, Italy Felipe P. Carpes Applied Neuromechanics Research Group, Faculty of Health Sciences, Federal University of Pampa, Uruguaiana, RS, Brazil Robert D. Catena Gait and Posture Biomechanics Lab, Washington State University, Pullman, WA, USA Andrea Cereatti Department POLCOMING, University of Sassari, Sassari, Italy Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, University of Sassari, Sassari, Italy Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy Elizabeth Chapman Workforce Solutions, BTE Technologies, Greenwood Village, CO, USA Sajal Chirvi Neuroscience Research Labs – Research 151, Medical College of Wisconsin, Zablocki VA Medical Center, Milwaukee, WI, USA
xxx
Contributors
Nachiappan Chockalingam Life Sciences and Education, Staffordshire University, Stoke On Trent, UK Swati Chopra Motion Analysis Laboratory, Mayo Clinic, Rochester, MN, USA Veronica Cimolin Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milano, Italy Daniel R. Clifton School of Health and Rehabilitation Sciences, The Ohio State University, Columbus, OH, USA Jameson Crane Sports Medicine Institute, The Ohio State University Wexner Medical Center, The Ohio State University, Columbus, OH, USA Brittany Coats Mechanical Engineering, University of Utah, Salt Lake City, UT, USA Mutlu Cobanoglu Department of Orthopedics and Traumatology, Adnan Menderes University Faculty of Medicine, Aydın, Turkey Scott Coleman Department of Orthopaedics, Baylor University Medical Center, Dallas, TX, USA Department of Orthopedics, Baylor Scott and White, Dallas, TX, USA John Collomosse Centre for Vision Speech and Signal Processing (CVSSP), University of Surrey, Surrey, UK Justin Connor Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Carmela Conte Movement Analysis LAB, Rehabilitation Centre Policlinico Italia, Rome, Italy Stelian Coros Carnegie Mellon University, Pittsburgh, PA, USA Emily S. Cross Bangor University, Bangor, North Wales, UK Robert G. Crowther Sport and Exercise, School of Health and Wellbeing, University of Southern Queensland, Ipswich, QLD, Australia Smart Movement, Brisbane, QLD, Australia Len Cubitt Tullamarine, VIC, Australia Andrea Giovanni Cutti Applied Research, INAIL Prosthetic Center, Vigorso di Budrio, BO, Italy Sofia Dahl Department of Architecture, Design and Media Technology, Aalborg University Copenhagen, Copenhagen, Denmark Annet Dallmeijer Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, EMGO+ Institute for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands
Contributors
xxxi
Jon R. Davids Northern California Shriner’s Hospital for Children, Sacramento, Sacramento, CA, USA Karen Davies Shriners Gait Lab, Sunny Hill Health Centre for Children, Vancouver, BC, Canada Alan R. De Asha C-Motion Inc., Germantown, MD, USA Ugo Della Croce Department POLCOMING, University of Sassari, Sassari, Italy Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, University of Sassari, Sassari, Italy Zhigang Deng Department of Computer Science, University of Houston, Houston, TX, USA Timothy R. Derrick Department of Kinesiology, Iowa State University, Ames, IA, USA Yu Ding University of Houston, Houston, TX, USA Andrew T. Duchowski Clemenson University, Clemson, SC, USA Raphaël Dumas LBMC UMR_T9406, Univ Lyon, Université Claude Bernard Lyon 1, IFSTTAR, Lyon, France Jan M. Eckerle Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany W. Brent Edwards Human Performance Laboratory, Faculty of Kinesiology, University of Calgary, Calgary, AB, Canada Arezoo Eshraghi Holland Bloorview Kids Rehabilitation Hospital, Bloorview Research Institute, Toronto, ON, Canada Stefano Federici Department of Philosophy, Social and Human Sciences and Education, University of Perugia, Perugia, Italy Anne M. Felts Workforce Solutions, BTE Technologies, Greenwood Village, CO, USA Malindu E. Fernando Podiatry Service, Kirwan Community Health Campus, Townsville, QLD, Australia College of Medicine, James Cook University, Townsville, QLD, Australia Maurizio Ferrarin Biomedical Technology Department, IRCCS Fondazione Don Carlo Gnocchi Onlus, Milan, MI, Italy Hilde Feys Research Group for Neuromotor Rehabilitation, KU Leuven, Leuven, Belgium
xxxii
Contributors
Henryk Flashner Department of Aerospace and Mechanical Engineering, University of Southern California, Los Angeles, CA, USA Peter G. Gabos Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Deborah Gaebler-Spira Shirley Ryan Ability Lab, Chicago, IL, USA Manuela Galli Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milano, Italy IRCCS “San Raffaele Pisana” Tosinvest Sanità, Roma, Italy Steven A. Gard Department of Physical Medicine and Rehabilitation, Feinberg School of Medicine, Northwestern University Prosthetics-Orthotics Center (NUPOC), Chicago, IL, USA Department of Physical Medicine and Rehabilitation, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA Jesse Brown VA Medical Center, Department of Veterans Affairs, Chicago, IL, USA Tom Gardner Bangor University, Bangor, North Wales, UK Ismat Ghanem Laboratory of Biomechanics and Medical Imaging, Faculty of Medicine, University of Saint-Joseph, Mar Mikhael, Beirut, Lebanon Hôtel-Dieu de France Hospital, University of Saint-Joseph, Beirut, Lebanon Claudia Giacomozzi Department of Cardiovascular Diseases, Dysmetabolic Diseases and Ageing, Italian National Institute of Health, Rome, Italy Werner Goebl Department of Music Acoustics – Wiener Klangstil (IWK), University of Music and Performing Arts Vienna, Vienna, Austria Brian D Goodwin Neuroscience Research Labs – Research 151, Medical College of Wisconsin, Zablocki VA Medical Center, Milwaukee, WI, USA Arnaud Gouelle Gait and Balance Academy, ProtoKinetics, Havertown, PA, USA Hans Gray Department of Mechanical Engineering, The University of Melbourne, Parkville, VIC, Australia Shanyuanye Guan Department of Mechanical Engineering, The University of Melbourne, Parkville, VIC, Australia Ikhsanul Habibie School of Informatics, University of Edinburgh, Edinburgh, UK Joseph Hamill Department of Kinesiology, University of Massachusetts, Amherst, MA, USA Andrew Hansen Minneapolis VA Health Care System, Minneapolis, MN, USA University of Minnesota, Minneapolis, MN, USA
Contributors
xxxiii
Aoife Healy Life Sciences and Education, Staffordshire University, Stoke On Trent, UK Laura A. Held Department of Biological Sciences, University of Southern California, Los Angeles, CA, USA John Henley Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Kasee J. Hildenbrand Athletic Training Program, Washington State University, Pullman, WA, USA Adrian Hilton Centre for Vision Speech and Signal Processing (CVSSP), University of Surrey, Surrey, UK Edmond S. L. Ho Department of Computer and Information Sciences, Northumbria University, Newcastle upon Tyne, UK At L. Hof Center for Human Movement Sciences and Laboratory of Human Movement Analysis, Department of Rehabilitation, University Medical Center Groningen, Groningen, The Netherlands Daniel Holden School of Informatics, University of Edinburgh, Edinburgh, UK H. Houdijk Faculty of Behavioral and Movement Sciences, Vrije Universiteit Amsterdam, Amsterdam Movement Sciences, Amsterdam, The Netherlands Heliomare Rehabilitation, Research and Development, Wijk aan Zee, The Netherlands Cheryl L. Hubley-Kozey School of Physiotherapy, Faculty of Health Professions, Dalhousie University, Halifax, NS, Canada School of Biomedical Engineering, Faculty of Medicine, Dalhousie University, Halifax, NS, Canada Nova Scotia Health Authority, Halifax, NS, Canada François Hug Laboratory “Movement, Interaction, Performance” (EA4334), University of Nantes, Nantes, France NHMRC Centre of Clinical Research Excellence in Spinal Pain, Injury and Health, School of Health and Rehabilitation Sciences, The University of Queensland, Brisbane, QLD, Australia Patria A. Hume Auckland University of Technology, Auckland, New Zealand T. IJmker Faculty of Behavioral and Movement Sciences, Vrije Universiteit Amsterdam, Amsterdam Movement Sciences, Amsterdam, The Netherlands Ellen Jaspers Neural Control of Movement Lab, ETH Zurich, Zurich, Switzerland Sophie Jörg School of Computing, Clemson University, Clemson, SC, USA
xxxiv
Contributors
Xiaogang Jin State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China Johanna Jonsdottir LaRiCE, Department of Neurorehabilitation, IRCCS Fondazione Don Carlo Gnocchi Onlus, Milan, Italy Justin Michael Kane Baylor University Medical Center, McKinney, TX, USA Faculty, Foot and Ankle Fellowship Program, Baylor University Medical Center, Dallas, TX, USA Orthopedic Associates of Dallas, Dallas, TX, USA Kenton R. Kaufman Motion Analysis Laboratory, Mayo Clinic, Rochester, MN, USA J. Keogh Bond University Australia, Robina, QLD, Australia Thomas M. Kepple C-Motion Inc., Germantown, MD, USA Matthew Klinker Workforce Solutions, BTE Technologies, Greenwood Village, CO, USA Isabella T. Klöpfer-Krämer Berufsgenossenschaftliche Unfallklinik Murnau und Paracelsus Medizinische Privatuniversität Salzburg, Institut für Biomechanik, Murnau am Staffelsee, Germany Taku Komura School of Informatics, University of Edinburgh, Edinburgh, UK Andrea Kotanxis Leon Root Motion Analysis Laboratory, Hospital for Special Surgery, New York, NY, USA Andreas Kranzl Laboratory for Gait and Human Motion Analysis, Orthopedic Hospital Speising, Vienna, Austria Eva G. Krumhuber University College London, London, UK Alberto Leardini Movement Analysis Laboratory and Functional-Clinical Evaluation of Prostheses, Istituto Ortopedico Rizzoli, Bologna, Italy Fabien Leboeuf School of Health Sciences, University of Salford, Salford, UK Nancy Lennon Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Lise Leveille Shriners Gait Lab, Sunny Hill Health Centre for Children, Vancouver, BC, Canada British Columbia Children’s Hospital, Vancouver, BC, Canada Glen Lichtwark Centre for Sensorimotor Performance, School of Human Movement and Nutrition Sciences, The University of Queensland, St Lucia, QLD, Australia Peter Loan C-Motion, Inc., Germantown, MD, USA Benjamin Lok Computer and Information Sciences and Engineering Department, University of Florida, Gainesville, FL, USA
Contributors
xxxv
William G. Mackenzie Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Michael W. Maier Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany Lisa Mailleux Research Group for Neuromotor Rehabilitation, KU Leuven, Leuven, Belgium Martina Mancini Department of Neurology, Oregon Health and Science University, Portland, OR, USA Susan Margulies Bioengineering, University of Pennsylvania, Philadelphia, PA, USA Carlo Massaroni Unit of Measurements and Biomedical Instrumentation, Campus Bio-Medico di Roma University, Rome, Italy Garrett A. Mattos Transport and Road Safety (TARS) Research Centre, University of New South Wales, Sydney, NSW, Australia Andrew S. McIntosh Australian Collaboration for Research into Injury in Sport and its Prevention (ACRISP), Federation University Australia, Ballarat, VIC, Australia Monash University Accident Research Centre, Monash University, Melbourne, VIC, Australia McIntosh Consultancy and Research, Sydney, NSW, Australia James McLoughlin Flinders University, Adelaide, Australia Jill L. McNitt-Gray Departments of Biological Sciences and Biomedical Engineering, University of Southern California, Los Angeles, CA, USA Fabio Meloni Department of Philosophy, Social and Human Sciences and Education, University of Perugia, Perugia, Italy Fabrice Mégrot Unité Clinique d’Analyse de la Marche et du Mouvement, Centre de Médecine Physique et de Réadaptation pour Enfants de Bois-Larris – CroixRouge Française, Lamorlaye, France UMR CNRS 7338: Biomécanique et Bioingénierie, Sorbonne Universités, Université de Technologie de Compiègne, Compiègne, France Freeman Miller Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Ross H. Miller Department of Kinesiology, University of Maryland, College Park, MD, USA Tomohiko Mukai Tokai University, Tokyo, Japan
xxxvi
Contributors
Robert Needham Life Sciences and Education, Staffordshire University, Stoke On Trent, UK Michael Neff Department of Computer Science and Program for Cinema and Digital Media, University of California – Davis, Davis, CA, USA Christopher Nester School of Health Sciences, University of Salford, Salford, UK Timothy A. Niiler Gait Laboratory, Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA Manfred Nusseck Freiburg Institute for Musicians’ Medicine, University of Music Freiburg, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany Sylvia Õunpuu Center for Motion Analysis, Division of Orthopaedics, Connecticut Children’s Medical Center, Farmington, CT, USA James A. Onate School of Health and Rehabilitation Sciences, The Ohio State University, Columbus, OH, USA Jameson Crane Sports Medicine Institute, The Ohio State University Wexner Medical Center, The Ohio State University, Columbus, OH, USA Michael Orendurff Motion and Sports Performance Laboratory, Lucile Packard Children’s Hospital Stanford, Sunnyvale, CA, USA Marcus Pandy Department of Mechanical Engineering, The University of Melbourne, Parkville, VIC, Australia Ilaria Parel Unit of Shoulder and Elbow Surgery, Cervesi Hospital, Cattolica, RN, Italy Dimitrios A. Patikas School of Physical Education and Sport Science, Aristotle University of Thessaloniki, Thessaloniki, Greece Declan A. Patton Australian Collaboration for Research into Injury in Sport and its Prevention (ACRISP), Federation University Australia, Ballarat, VIC, Australia Oslo Sports Trauma Research Centre (OSTRC), Norwegian School of Sport Sciences, Oslo, Norway Sport Injury Prevention Research Centre (SIPRC), University of Calgary, Calgary, AB, Canada Catherine Pelachaud CNRS - ISIR, Université Pierre et Marie Curie, Paris, France Kristan Pierz Center for Motion Analysis, Division of Orthopaedics, Connecticut Children’s Medical Center, Farmington, CT, USA Frank A. Pintar Neuroscience Research Labs – Research 151, Medical College of Wisconsin, Zablocki VA Medical Center, Milwaukee, WI, USA
Contributors
xxxvii
Jessica May Pohlmann Sport and Exercise, University of Southern Queensland, Ipswich, QLD, Australia Smart Movement, Brisbane, Australia Ana Presedo Pediatric Orthopaedics Department, Robert Debré University Hospital, Paris, France Jessica Pruente Shirley Ryan Ability Lab, Chicago, IL, USA Ilona M. Punt Department of Epidemiology, Maastricht University, CAPHRI, Maastricht, The Netherlands Department of Physical Therapy, University of Applied Sciences of Western Switzerland, Carouge, Switzerland Martin Puttke Board of German Federal Association of Dance, Berlin, Germany Elizabeth A. Rapp University of Delaware, Newark, DE, USA Julie Reay School of Health Sciences, University of Salford, Salford, UK Jiaping Ren State Key Lab of CAD&CG, Zhejiang University, Hangzhou, China Jaques Riad Skaraborg Hospital Skövde, Skövde, Sweden André Gonçalo Gomes Roque Physiotherapy, University of Averio, Aveiro, Portugal Marlene Cristina Neves Rosa Piaget Institute, Viseu, Portugal Dieter Rosenbaum Funktionsbereich Bewegungsanalytik, Institut für Experimentelle Muskuloskelettale Medizin, Zentrum für Muskuloskelettale Medizin, Universitätsklinikum Münster, Münster, Germany Bodo Rosenhahn Institut für Informationsverarbeitung, Leibniz Universität Hannover, Hannover, Germany Rüdiger Rupp Spinal Cord Injury Center – Experimental Neurorehabilitation, Heidelberg University Hospital, Heidelberg, Germany Erich Rutz Pediatric Orthopaedic Department, University Children’s Hospital Basel, Basel, Switzerland Angelo M. Sabatini The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy Najmeh Sadoughi Multimodal Signal Processing Lab, University of Texas at Dallas, Dallas, TX, USA Ross H. Sanders Exercise and Sport Science, Faculty of Health Sciences, The University of Sydney, Sydney, NSW, Australia Morgan Sangeux Hugh Williamson Gait Analysis Laboratory, The Royal Children’s Hospital, Parkville/Melbourne, VIC, Australia
xxxviii
Contributors
Gait laboratory and Orthopaedics, The Murdoch Childrens Research Institute, Parkville/Melbourne, VIC, Australia Andrea Schärli Institute of Sport Science, University of Bern, Bern, Switzerland Jonathan Schwarz School of Informatics, University of Edinburgh, Edinburgh, UK W. Scott Selbie HAS-Motion Inc., Kingston, ON, Canada C-Motion Inc., Germantown, MD, USA Mariano Serrao Department of Medical and Surgical Sciences and Biotechnologies, Sapienza University of Rome, Latina, Italy Movement Analysis LAB, Rehabilitation Centre Policlinico Italia, Rome, Italy Gongbing Shan Department of Kinesiology, Faculty of Arts and Science, University of Lethbridge, Lethbridge, AB, Canada Frances T. Sheehan Rehabilitation Medicine Department, Functional and Applied Biomechanics Section, National Institutes of Health, Bethesda, MD, USA Yijun Shen Northumbria University, Newcastle upon Tyne, UK Fuhao Shi Texas A&M University, College Station, TX, USA Andrew Short University of Melbourne, Melbourne, Australia Adam Shortland One Small Step Gait Laboratory, Evelina Children’s Hospital, Guy’s and St Thomas’ NHS Foundation Trust, London, UK Hubert P. H. Shum Northumbria University, Newcastle upon Tyne, UK Anne K. Silverman Functional Biomechanics Laboratory, Department of Mechanical Engineering, Colorado School of Mines, Golden, CO, USA Henry M. Silvester Barry Nilsson Lawyers, Sydney, NSW, Australia Cristina Simon-Martinez Research Group for Neuromotor Rehabilitation, KU Leuven, Leuven, Belgium Wafa Skalli Institut de Biomécanique Humaine Georges Charpak, Arts et Métiers ParisTech, Paris, France Lina Skora University College London, London, UK L. H. Sloot Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, VU University Medical Center, Amsterdam, The Netherlands Colin R. Smith Department of Mechanical Engineering, University of WisconsinMadison, Madison, WI, USA Richard M. Smith Rehabilitation Medicine Department, Functional and Applied Biomechanics Section, National Institutes of Health, Bethesda, MD, USA
Contributors
xxxix
Pierpaolo Sorrentino Department of Engineering, University of Naples Parthenope, Naples, Italy Giuseppe Sorrentino Department of Motor Sciences and Wellness, University of Naples Parthenope, Naples, Italy Institute Hermitage-Capodimonte, Naples, Italy Jacob J. Sosnoff Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL, USA Claudia Spahn Freiburg Institute for Musicians’ Medicine, University of Music Freiburg, Medical Center – University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany David M. Spranz Clinic for Orthopedics and Trauma Surgery, Heidelberg University Hospital, Heidelberg, Germany Felix Starker Biomechatronic Systems, Fraunhofer Institute for Manufacturing Engineering and Automation, Stuttgart, Germany Julie A. Stebbins Oxford Gait Laboratory, Oxford University Hospitals NHS Foundation Trust, Oxford, UK Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, Oxford, UK Felix Stief Movement Analysis Lab, Orthopedic Friedrichsheim gGmbH, Frankfurt/Main, Germany
University
Hospital
Ruopeng Sun Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL, USA Sean A. Tabaie Northern California Shriner’s Hospital for Children, Sacramento, Sacramento, CA, USA Hideki Takagi Faculty of Health and Sport Sciences, University of Tsukuba, Ibaraki, Japan Tomoichi Takahashi Department of Information Engineering, Meijo University, Nagoya, Japan Jie Tan Georgia Institute of Technology, Atlanta, GA, USA Darryl G. Thelen Department of Mechanical Engineering, University of Wisconsin-Madison, Madison, WI, USA Pam Thomason Hugh Williamson Gait Analysis Laboratory, Royal Children’s Hospital, Melbourne, VIC, Australia Kylie Tucker NHMRC Centre of Clinical Research Excellence in Spinal Pain, Injury and Health, School of Health and Rehabilitation Sciences, The University of Queensland, Brisbane, QLD, Australia
xl
Contributors
School of Biomedical Sciences, The University of Queensland, Brisbane, QLD, Australia Brian R. Umberger Department of Kinesiology, University of Massachusetts, Amherst, MA, USA Michiel van de Panne University of British Columbia, Vancouver, BC, Canada M. M. van der Krogt Department of Rehabilitation Medicine, MOVE Research Institute Amsterdam, VU University Medical Center, Amsterdam, The Netherlands Giuseppe Vannozzi Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome Foro Italico, Rome, Italy Benedicte Vanwanseele Department of Movement Sciences, KU Leuven, Leuven, Belgium Fontys University of Applied Sciences, Eindhoven, The Netherlands Mikko Virmavirta Biology of Physical Activity, The Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland Peter Visentin Department of Music, Faculty of Fine Arts, University of Lethbridge, Lethbridge, AB, Canada Dimitri Volchenkov Mathematics and Statistics, Texas Tech University, Lubbock, TX, USA Center for Nonlinear Physics, Sichuan University of Science and Engineering, Sichuan, China J. J. Wallace Motion Analysis Center, Shriners Hospitals for Children Medical Center, Lexington, KY, USA Marcelo M. Wanderley Input Devices and Music Interaction Laboratory (IDMIL), CIRMMT, McGill University, Montreal, QC, Canada Bastian Wandt Institut für Informationsverarbeitung, Leibniz Universität Hannover, Hannover, Germany Lijuan Wang Microsoft Research, Redmond, WA, USA Scott Wearing Institute of Health and Biomedical Innovation, Queensland University of Technology, Brisbane, QLD, Australia Faculty for Sport and Health, Technische Universität München, Munich, Bavaria, Germany Sarah Whatley Centre for Dance Research, Coventry University, Coventry, UK Hank White Motion Analysis Center, Shriners Hospitals for Children Medical Center, Lexington, KY, USA
Contributors
xli
Steffen Willwacher Institute of Biomechanics and Orthopedics, German Sport University, Cologne, Germany Institute of Functional Diagnostics, Cologne, Germany Janis Wojtusch Department of Computer Science, Simulation, Systems Optimization and Robotics Group, TU Darmstadt, Darmstadt, Germany Tyler A. Wood Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL, USA Lei Xie School of Computer Science, Northwestern Polytechnical University (NWPU), Xi’an, P. R. China Shan Yang School of Computer Science, Northwestern Polytechnical University, Xi’an, China Feng Yang Department of Kinesiology, The University of Texas at El Paso, El Paso, TX, USA Longzhi Yang Northumbria University, Newcastle upon Tyne, UK Maurice R. Yeadon Loughborough University, Loughborough, UK KangKang Yin Simon Fraser University, Burnaby, BC, Canada Department of Computer Science, Singapore, Singapore Bing Yu Division of Physical Therapy, Department of Allied Health Science, School of Medicine, The University of North Carolina at Chapel Hill, Chapel Hill, NC, USA Pong C. Yuen Department of Computer Science, Hong Kong Baptist University, Kowloon Tong, Hong Kong Petrissa Zell Institut für Informationsverarbeitung, Leibniz Universität Hannover, Hannover, Germany Jingtian Zhang Northumbria University, Newcastle upon Tyne, UK
Part I Rigid Body Modeling
Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries Aurelio Cappozzo
Abstract
Observing, revealing the hidden structure, and understanding the human locomotor system have been a goal for artists alone at first and for both artists and scientists later. How and why this goal was achieved is illustrated in this chapter. This is done following the fil rouge of history, believing that the understanding of the phylogenesis of knowledge effectively accompanies its ontogenesis. The realistic representation of the human form, as opposed to its metaphorical depiction, started in ancient Greece, but reached its apex in the Renaissance when artists understood that they needed a deeper understanding of reality in order to create an illusion of it. This was the premise for the scientific revolution in general and with regard to human motion in particular. During the nineteenth century, thermodynamics and the introduction of novel measurement and recording technologies gave renewed impulse to the study of the human locomotor system as if it were a machine designed either for fighting or for working. At the beginning of the twentieth century, avant-garde artists cooperated with the science of human movement by adding stronger human emotions and feelings to the scientific narration and therefore establishing a deeper perception of the natural phenomenon. The development of reconstructive orthopedic surgery in the second half of the twentieth century made the acquisition of deeper knowledge about the hidden movement of bones and the mechanics of human joints urgent. Since then, ever more sophisticated and accurate mathematical models of the neuromusculoskeletal system have been developed taking advantage of digital technology for measurement, computing, and virtual animation.
A. Cappozzo (*) Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome “Foro Italico”, Rome, Italy e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_199
3
4
A. Cappozzo
Keywords
History • Human movement • Locomotor apparatus • Biomechanics
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Throughout the Centuries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Until the Middle Ages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Renaissance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . From the Age of Enlightenment to the Industrial Revolution or the Machine Age . . . . . . . . . Toward the Digital Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Contemporary Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 4 4 6 7 10 13 14 14
Introduction This chapter tackles a very basic problem in the biomechanics of the human locomotor apparatus. It’s about the observation – in the Galilean sense of course – of the human form while it moves, the observation of what is visible and, above all, of what is not visible, that is, the hidden structure of the locomotor system. To this purpose the fil rouge of history will be followed, believing that the understanding of the phylogenesis of knowledge effectively accompanies the ontogenesis of it. The reader will therefore be accompanied through the different methods and purposes that characterized the observation of human motion in the course of the past centuries and as far as the threshold of the current state of the art presented in this handbook.
Throughout the Centuries Until the Middle Ages For several millennia, the observation of a natural phenomenon, in general, and of the human form in motion, in particular, has been the prerogative of artists and philosophers in terms of the unveiling of its hidden structure, its interpretation, and its reproduction either in a symbolic or in a realistic fashion. As paradigmatic examples in this respect, we may examine two masterpieces that we have inherited from the ancient Greeks. One is represented by a Kouros (seventh century before the Christian era, Fig. 1). This is a statue depicting a young man who may seem to walk, but obviously goes nowhere. This was not due to inability of the artist to depict reality, quite the contrary, since the objective was not conveying a message of mobility, of life, but it was portraying a timeless figure, an image that transcends time. This was appropriate because this figure was a grave marker. The second
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
5
Fig. 1 Statue of a Kouros (young man) (580 BCE). New York, The Metropolitan Museum of Art
example is the Doryphoros by Polykleitos (440 before the Christian era, Fig. 2). Here the artist observed the outer form of the body, guessed the muscular and bony structure beneath the skin, and produced a figure that exists in the real world and is moving in it. This figure displays the correct pelvic attitude and the correct hip, knee, and ankle angles. All six determinants of gait, as described by Saunders and colleagues in 1953 are there! The Kouros and the Doryphoros represent two different ways of modelling reality. Just like our mathematical models, which may be metaphorical or naturalistic. Ancient Rome was a very pragmatic interpreter of Greek as well as Etruscan, native Italic, and even Egyptian visual culture, but did not bring any novelty in the way the human form, stationary or in motion, was represented. Medieval images of the human body were characterized by the fact that they did not reflect close observation from real life. Again, they aimed at representing an idea, and they also were metaphorical representations in nature. In those times, in fact, at least in the western part of the world, there was very little interest in the human body and in its inner workings. The body was seen as a mere temporary receptacle of the soul and cause of temptation.
6
A. Cappozzo
Fig. 2 A Roman copy of The Doryphoros (spear bearer) by Polykleitos (440 BCE). Naples, The National Archaeological Museum
The Renaissance Naturalistic representation is retrieved and reached its apex in the Renaissance (fourteenth to the seventeenth century) when artists realized that it was not sufficient to observe only the outer surface of a human body, as the ancient Greeks did, but it was necessary to go inside it and study anatomy through dissection. This approach allowed them to paint or sculpt the outer appearance of the body in many different static and dynamic postures accounting for muscle shape variations due to contraction, tension in the tendons, bony prominences, etc. This entailed revealing the outer human form starting from the hidden structure underneath, what we may name an “inside-outward” approach. Adam’s body as depicted in the Sistine Chapel in Rome is the mighty result of Michelangelo’s dissections and anatomical observations. Renaissance artists understood that they needed a deep understanding of reality in order to create an illusion of it. This awareness shows that times were ripe for an artist or philosopher to slowly and progressively start changing their skin and undergo a mutation and eventually become a scientist or rather a “natural philosopher.” At this time, art and science started a common itinerary and became travelling companions toward the ultimate goal of understanding, possessing, and controlling
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
7
natural phenomena, among which the human body, stationary or in motion, made no exception. Another important event characterized the Renaissance: mathematics sneaked into the artists’ work and became as important as observation. Linear perspective and the mathematics associated with it were invented. The three-dimensional world could be effectively portrayed on a planar canvas or wall. This is the same mathematics we use today in analytical stereophotogrammetry, although the other way round, using planar views, we reconstruct the object of interest in its three dimensions. The anthropometric measurements of Leonardo da Vinci (1510) as well as of Albrecht Durer (1528) may also be framed in this mathematical context. Of course, the ancient Greeks had already touched upon this last endeavor, as shown by the golden ratio or divine proportion that has intrigued artists, architects, philosophers, and scientists from Euclid onward. The Roman architect Vitruvius Polonius, who inspired the famous Leonardo’s drawing, named the Vitruvian Man (Stemp 2006), should also be mentioned in this context. In the seventeenth century, quantitative experimental observation, coupled with the analytical power of mathematics, became the normal approach for the purpose of describing and understanding the natural phenomenon and foreseeing or forecasting its evolution in time and space. This is the so-named scientific revolution. At that time, Galileo’s message struck a young medical doctor whose name was Giovanni Alfonso Borelli. He was the first one to apply the scientific method to the study of human motion, but with two important limitations: first, he was not aware of the inertia principle, thus his analyses were strictly static, and, second, he had no adequate instrumentation available. In addition, as he himself admitted, he was unable to carry out anatomical dissections. Despite this, his book, the De Motu Animalium, remains an unparalleled masterpiece (Borelli 1680, 1681). On purely speculative grounds, Borelli described human walking using the paradigm of the compass familiar to all human movement scientists. He also carried out the first estimate of internal loads as depicted in Fig. 3. Borelli added up the magnitudes of the forces exerted by single muscles and concluded that an individual must exert a total muscular force that is almost 50 times larger than the transported load. A rather naïve procedure, of course, but the message is strong and effective (Cappozzo and Marchetti 1992).
From the Age of Enlightenment to the Industrial Revolution or the Machine Age The abovementioned scientific revolution finds its natural continuation in the Age of Enlightenment that characterized the eighteenth century. Although this was an outstanding and absolutely crucial fragment of the history of mankind, no notable scientific contribution to the understanding of human motion can be recorded. It is, however, worth mentioning what Diderot and d’Alembert wrote in their encyclopedia (1751–1772) about the notion of “movement” as applied to a human being: “Movement is the action of a living body which is necessary for the conservation of
8
A. Cappozzo
Fig. 3 Frontispiece of Borelli’s De Motu Animalium first volume. External and internal forces acting on the human body while carrying a load
its health; the lack of movement as well as the excess of it are extremely prejudicial to the body.” During the nineteenth century, the formalization of thermodynamics gave renewed impulse to the study of the human locomotor system looked upon as if it were a machine designed either for fighting, in the rather unsettled Europe of that time, or for working, in the Second Industrial Revolution framework. It is not by chance that books reporting studies on human movement were given titles such as The Animal Machine (Marey 1873) or The Human Motor (Amar 1914). Of course technology helped this endeavor: Marey’s sensors and recording devices (Marey 1885, 1894), Muybridge’s highly sensitive photographic material (Muybridge 1887, 1899, 1901; Stillman 1882), Braune and Fischer’s stereophotogrammetric model (Braune and Fischer 1895–1904), and, later on, the high sampling rates used by Bernstein to record movement, just to mention the most important achievements. For a review of the work of these scientists, the reader may refer to Bouisset (1992), Tosi (1992), Maquet (1992), and Jansons (1992), respectively. At the beginning of the twentieth century, art outshined the science of human movement adding stronger human emotions and feelings to the scientific narration and therefore a deeper perception of the natural phenomenon. This, in turn, means a more thorough knowledge, a humanistic knowledge of the phenomenon that goes well beyond what may be stored in the memory of a computer. The “Nude Descending a Staircase”, painted by Marcel Duchamp in 1912, is a vivid example in this respect. Another example worth mentioning is the Italian artistic and philosophical movement named Futurism that celebrated the modern world of industry and technology, conjugated physics and aesthetics, and very effectively portrayed motion and speed both on canvas and in bronze through an evident influence of the abovementioned biomotion studies (Poore 1913; Fig. 4).
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
9
Fig. 4 Movement and dynamism of Futurism: (a) “Girl Running on a Balcony” by Giacomo Balla (1912), (b) “Unique Forms of Continuity in Space“ by Umberto Boccioni (1913)
In the course of the first half of the twentieth century, movement scientists could reconstruct the movement of a stick model of the human locomotor system as projected on a 2-D space or, in rare cases, in the 3-D space. These models were activated by the trajectories of target points located on the skin surface – most of the time on points approximating the joint centers – so that a sort of virtual monodimensional exoskeleton in motion could be reconstructed. However, no observation of any inner structure was attempted. These models, despite the obvious limitations and inaccuracies and the fact that they allowed more narrative than quantitative descriptions of the phenomena involved, sufficed the purpose of describing the human machine for pursuing several practical objectives. The impact that these studies had on society appears evident, for instance, when we look at the development of industrial production methodology. Just think of the Taylorism and Fordism and the implementation of these theories in America, where they were formulated, but also in Europe and even in the Soviet five-year plans. Human movement science, together with Pavlov’s learning theory, had a tremendous impact on performing arts as well. This is particularly evident in the work of the Russian theatrical actor and director Vsevolod Meyerhold who promoted an acting style, named Biomechanics, according to which “. . . it is when the actor has found the correct positions that he can pronounce the words, and only then these will sound meaningful . . ..” This acting style resulted to be in opposition with the more, so to speak, naturalistic Stanislavsky system (Law and Gordon 1995).
10
A. Cappozzo
Toward the Digital Age At the end of the Second World War, a very thorough study on human locomotion was carried out at the University of California with the aim of designing advanced lower limb substitutes (Eberhart 1947; Paul 1992). This study provided a piece of information regarding the description of human locomotion not available before. Specifically, the rotation of the pelvis, femur, and tibia about their longitudinal axes during walking was measured. For the first time we may legitimately talk of bone movement as opposed to segmental movement. But to this purpose pins had to be inserted into the volunteer’s bones acknowledging the fact that targets located on the skin surface would have not been able to track the rotation of the underlying bone reliably (Fig. 5). The development of endo- and arthro-prostheses and of reconstructive orthopedic surgery in general in the 1950s and 1960s made both the acquisition of deeper knowledge concerning the hidden movement of bones and the mechanics of human joints urgent. John Paul, at the University of Glasgow, responded to this urgency by initiating pioneering work unveiling the hidden structure around 1965. This author
Fig. 5 Test subject with pins and markers attached. Pelvis, femur, and tibia rotation during walking (Modified from Eberhart 1947)
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
11
used an outside-inward approach. The hidden structure was estimated using information collected on the body surface using cinecameras (Fig. 6). He was able to provide very useful data concerning the muscular forces and hip loading (Paul 1966, 1967, 1969), definitely a milestone in the history of biomechanics. At this point in time, we may say that two methodological approaches to the study of human movement are possible. While analyzing a motor task, one may aim at describing segmental movement using low to medium resolution (chapters ▶ “3D Dynamic Pose Estimation from Marker-Based Optical Data” and ▶ “3D Dynamic Pose Estimation from Markerless Optical Data”), and this may be adequate for a number of clinical applications such as motor function limitation assessment or in ergonomics and sports. The other approach leads to a high-resolution reconstruction of the movement of the inner invisible structures for the sake of musculoskeletal system modelling (chapters ▶ “Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and Ligament,” ▶ “Physics-Based Models for Human Gait Analysis,” and ▶ “Optimal Control Modeling of Human Movement”). When we talk of high resolution of bone pose reconstruction, we mean to be able to resolve millimeters and degrees, as paradigmatically illustrated in Iwaki et al. (2000) with respect to the knee joint. From the mid-1970s onward, methods have been proposed that in fact allow the direct observation of the bones and thus an adequate resolution for the purpose
Fig. 6 (a) Positions of markers on the test subject. (b) Typical curve of hip joint force to body weight ration with time (Modified from Paul 1967)
12
A. Cappozzo
Fig. 7 A set of Cartesian set of axes is associated with the 3-D marker (bone-embedded frame). (a) Estimating the instantaneous pose of a bone using a 3-D marker assumed to be rigidly associated with the bone; (b) the 3-D marker may be represented by a cluster of landmarks or (c) a magnetoinertial measurement unit
mentioned previously. One is roentgen-stereophotogrammetry, which is of course invasive and suitable only to assess micromotion of orthopedic implants or in ex vivo joint mechanics investigations (Selvik 1974, 1990). Another powerful method able to allow the direct observation of bones during motion and less prone to criticisms from the ethical point of view is digital fluoroscopy which started to be used in the present context at the end of the 1990s, thanks to the work of Scott Banks and Hodge (1996), and has come to full maturation in the course of the last decade (chapter ▶ “Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography”). However, it exhibits these limitations: it is invasive (ionizing radiation), the measurement volume is limited, and the procedure is operator time-consuming. In any case these limitations make the direct observation of bones in motion inapplicable when the objective is the analysis of large portions of the musculoskeletal system moving in large volumes and/or characterized by high accelerations. Under these circumstances other methods must be devised. These consist in the use of observable 3-D markers, that is, three-dimensional objects – external to the body, for this reason directly observable using noninvasive methods – which are assumed to be rigidly associated with the underlying bone (Fig. 7a) (chapter ▶ “Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units”). The abovementioned 3-D marker may be defined by three or more landmarks (Fig. 7b). The position in space of these landmarks may be recorded using retroreflective or light-emitting point markers located on the surface of the body segment
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
13
and stereophotogrammetry. As it occurred just about 100 years before with photography, the availability of optoelectronic sensors and high-performance digital technology at an accessible cost revolutionized experimental movement analysis. In the 1970s, time became ripe for the development of optoelectronic stereophotogrammetric systems. John Paul, by now at the Bioengineering Unit of Strathclyde University, understood it and invested the Ph.D. students Mick Jarrett and Brian Andrews with the responsibility of tackling this bold endeavor (Jarret et al. 1976). A possible alternative is using, as 3-D marker, a magneto-inertial measurement unit mounted on the body segment of interest (Fig. 7c) (chapter ▶ “Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units”). This technique was first explored in Oxford (UK) by the graduate student Julian Morris, under the supervision of John O’Connor (Morris 1973), using only accelerometers. Magneto-inertial measurement units have recently known a remarkable technological development in terms of miniaturization and performance. The signals provided by these devices, however, allow the estimate of the orientation of the bone-embedded frame, but not of its position. This circumstance of course limits, but does not exclude, their use in the present context. Due to the interposed soft tissues, the abovementioned assumption that the 3-D markers are rigid with the underlying bone is of course disputable, and this causes remarkable inaccuracies that must be taken care of. This is the so-named soft tissue artefact issue, the solution of which does not seem to be just round the corner (chapter ▶ “3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers”).
The Contemporary Age At the beginning of the 1980s, the above-illustrated optoelectronic stereophotogrammetric systems started to be marketed together with the relevant software. This was preceded by the appearance on the market of the first professional grade six-component force plate as a result of a joint venture between Jürg Wartenweiler of the ETH Zurich and a private company. At this point in time, we may say that the modern human movement analysis laboratory was finally available and the contemporary history of this discipline began. During the last decade the abovementioned laboratory has undergone a remarkable technological development and price reduction. Micro-Electro-Mechanical Systems and wireless technology have made wearable magneto-inertial sensors available almost ubiquitously, video-based markerless technology has become feasible, and medical imaging instrumentation is evolving rapidly, enriching the capability of observing both the morphology and the motion of inner structures while minimizing invasiveness. Lastly, methodologies developed in robotic engineering have lately shown that a cooperative synergy between this discipline and biomechanics is a very productive endeavor.
14
A. Cappozzo
Conclusion The evolution of human motion studies, throughout the different cultural ages of mankind, and the theoretical and applied motivations that drove it have been illustrated. This was intended to contribute to the understanding of the scenario within which the results presented in the subsequent chapters have been obtained and that forms the basis for future directions.
References Amar J (1914) Le Moteur Humainet les Bases Scientifiques du Travail Professional. Published by H. Dunod and E. Pinat, Paris Banks SA, Hodge WA (1996) Accurate measurement of three-dimensional knee replacement kinematics using single-plane fluoroscopy. IEEE Trans Biomed Eng 43(6):638–649 Borelli AJ (1680–1681) De Motu Animalium. Pars prima and Pars altera. Pub. by A. Bernabò, Roma. Translation by P. Maquet: On the movement of animals. Springer, Berlin/Heidelberg, 1989 Bouisset S (1992) Etienne-Jules Marey, or when motion biomechanics emerged as a science. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 71–88 Braune W, Fischer O (1895–1904) Der Gang des Menschen. Published by B.G. Teubner. Translation by P. Maquet and R: Furlon: The human gait. Published by Springer, Berlin/Heidelberg, New York, London, Paris, Tokyo, 1987 Cappozzo A, Marchetti M (1992) Borelli’s heritage. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 33–47 Leonardo da Vinci (1510) Human proportions. In: Anatomical drawings, Royal Library – Windsor Castle, pp 143–147. http://www.metmuseum.org/art/metpublications/leonardo_da_vinci_ana tomical_drawings_from_the_royal_library_windsor_castle# Dürer A (1528) VierBücher von Menschlicher Proportion (Four Books on Human Proportion). Published by Hieronymus Formschneyder, Nuremberg Eberhart HD (1947) Fundamental studies of human locomotion and other information relating to design of artificial limbs. Prosthetic Devices Research Report. University of California, Berkeley Iwaki H, Pinskerova V, Freeman MA (2000) Tibiofemoral movement: the shapes and relative movements of the femur and tibia in the unloaded cadaver knee. J Bone Joint Surg Br 82 (8):1189–1195 Jansons H (1992) Bernstein: the microscopy of movement. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 137–174 Jarret MO, Andrews BJ, Paul JP (1976) A television/computer system for the analysis of human locomotion. In: IERE golden jubilee conference on the applications of electronics in medicine. IERE Conference Proceedings No. 34 JBdecM S, Inman VT, Eberhart HD (1953) The major determinants in normal and pathological gait. J Bone Joint Surg Am 35:543–558 Law AH, Gordon M (1995) Meyerhold, Eisenstein and biomechanics: actor training in revolutionary Russia. Published by McFarland & co, Jefferson Maquet P (1992) “The human gait” by Braune and Fischer. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 115–126
Observing and Revealing the Hidden Structure of the Human Form in Motion. . .
15
Marey E-J (1873) La Machine Animale, Locomotion TerrestreetAérienne. Published by G. Baillie, Paris Marey E-J (1885) La Méthode Graphique dans les Sciences Expérimentales. Published by G. Masson, Paris Marey E-J (1894) La Méthode Graphique dans les Sciences Expérimentales. Second édition avec supplément. Le Développement de la Méthode Graphique par la Photographie. Published by G. Masson, Paris Morris JRW (1973) Accelerometry – a technique for the measurement of human body movements. J Biomech 6(6):729–732 Muybridge E (1887) Animal Locomotion: an electro-photographic investigation of consecutive phases of animal movements, commenced 1872 – completed 1885. Published under the auspices of the University of Pennsylvania by J.B. Lippincott Co Muybridge E (1899) Animals in Motion. Published by Chapman and Hall (reprinted by Dover, New York, 1957) Muybridge E (1901) The human figure in motion. Published by Chapman and Hall (reprinted by Dover, New York, 1955) Paul JP (1966) Biomechanics. The biomechanics of the hip-joint and its clinical relevance. Proc R Soc Med 59(10):943–948 Paul JP (1967) Forces at the human hip joint. PhD thesis. http://theses.gla.ac.uk/3913/ Paul JP (1969) Loading on the head of the femur. J Anat 105:187–188 Paul JP (1992) The Californian contribution. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 176–195 Poore, H.R., 1913. The new tendency in art: post impressionism, cubism, futurism. Published by Cornell University Library's print collections and scanned on an APT BookScan and converted to JPG 2000 format by Kirtas Technologies in 2009 Selvik GA (1974) Roentgen stereophotogrammetric method for the study of the kinematics of the skeletal system. Ph.D. thesis, University of Lund, Sweden Selvik GA (1990) Roentgen stereophotogrammetric analysis. Acta Radiologica 31(2):113–126 Stemp, R. (2006) The secret language of the Renaissance: decoding the hidden symbolism of Italian Art. Published by Duncan Baird, London Stillman JDB (1882) The horse in motion as shown by instantaneous photography, with a study on animal mechanics founded on anatomy and the revelations of the camera, in which is demonstrated the theory of quadrupedal locomotion. Published by J. R. Osgood & Co., Boston Tosi H (1992) Marey and Muybridge: how modern biolocomotion analysis started. In: Cappozzo A, Marchetti M, Tosi V (eds) Biolocomotion: a century of research using moving pictures. Promograph, Roma, pp 51–70
Three-Dimensional Reconstruction of the Human Skeleton in Motion Valentina Camomilla, Aurelio Cappozzo, and Giuseppe Vannozzi
Abstract
This chapter illustrates the conceptual background underlying the in silico reconstruction of the human skeletal motion. A specific focus is given to the experimental and analytical methods that allow acquiring information related to both bone movement and morphology in vivo in the framework of rigid body mechanics. This process involves the definition of global and local frames of reference. Common anatomical and mathematical conventions that are used to describe global bone pose and joint kinematics are illustrated. Issues concerning accuracy and reliability of the estimated quantities when using skin markers and stereophotogrammetry and magneto-inertial measurement units are also dealt with. Keywords
Rigid body mechanics • Human movement analysis • Bone pose estimation • Anatomical calibration • Joint kinematics
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Global and Local Frames: Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Global Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Local Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18 18 22 22 24
V. Camomilla (*) • A. Cappozzo Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome “Foro Italico”, Rome, Italy e-mail: [email protected]; [email protected] G. Vannozzi Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, Department of Movement, Human and Health Sciences, University of Rome Foro Italico, Rome, Italy e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_146
17
18
V. Camomilla et al.
Estimate of the Bone-Embedded Frame Pose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stereophotogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magneto-Inertial Measurement Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anatomical Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Subject-Specific Morphology Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Movement-Morphology Data Registration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Construction of the Anatomical Frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joint Kinematics Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Translational Degrees of Freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rotational Degrees of Freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Precision and Accuracy of Joint Kinematics Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25 26 28 28 28 31 33 33 36 36 38 39 41
Introduction Protecting and enhancing human motor function is an important strategic aim of modern society dwelling within the grand challenges concerning health and wellbeing (personalized care, prevention and early diagnosis, integrative and holistic approaches, healthy aging, sustainability of chronic diseases). Efforts must be made to strengthen our multi-scale understanding of both structure and function of the human locomotor system and to develop investigative and operational methods that can be translated into professional practice. In this context, quantitative human movement analysis is of paramount importance, and, for many applications, it must be carried out in the three-dimensional space. This analysis is based on measurements (motion capture, dynamometry, electromyography, calorimetry, medical imaging, etc.) and computational modeling (mathematical models of the anatomy and physiology of the tissues, organs, and systems involved). It provides information on the functions of the locomotor subsystems and on the overall strategy of motor activity. These outcomes contribute to the understanding of the key factors that affect joint motion and internal loading and, thus, injury, tissue degeneration or regeneration, as well as motor control and its adaptation, energy consumption, and fatigue. Quantification of subjectspecific variables can be effectively used in prevention, early diagnosis (e.g., monitoring of functional status in the elderly, specific workers, or athletes), intervention (e.g., prognosis, therapeutic programming, workplace optimization, training), and quantifying relevant outcomes. Prospectively, advanced movement analysis technology may be used for the implementation of real-time biofeedback (virtual and augmented reality) both in rehabilitation and training (institutionalized or not).
State of the Art A prerequisite for many of the above-mentioned endeavors is the accurate threedimensional reconstruction in silico of the portion of the locomotor system of interest while performing a motor task. The solution of this problem is based on the following fundamental considerations.
Three-Dimensional Reconstruction of the Human Skeleton in Motion
19
A body may be thought to be made of P particles. The larger the number P, the more details of the body shape are portrayed. The motion of this body may be described by providing information about the motion of each of those particles in the form, for instance, of three Cartesian coordinates (or a position vector), relative to a given set of orthogonal axes (global frame), and their time derivatives, in each sampled instant of time (1. . .N ). The resulting set of numbers describing position is made of 3 P N numbers. This data set may be split into two parts: one describing movement and the other describing shape. This is a profitable exercise because, if we are not interested in shape variation, i.e., in deformation, the second data subset may be considered to be time invariant, and this results in a remarkable reduction of the data set dimension with the obvious experimental and computational advantage. This is achieved through the following procedure. A set of orthogonal axes, generally named local frame (l), is defined rigidly associated with the body while assuming its current shape (Fig. 1). The location in space of this frame relative to the global frame (g), in a given instant of time, is described using two separate bits of information: one deals with its position and the other with its orientation (also named attitude). The former information is provided by the position vector of the origin of the local frame (Fig. 1): g
o¼
g
ox g oy g oz
(1)
The orientation of the local frame may be described by providing the orientation of each of its axes. This is done through the components of the three unit vectors aligned with these axes, relative to the global frame axes. Through obvious geometrical considerations, it can be seen that these components are equal to the cosines (named direction cosines) of the angles formed by each unit vector with the global frame axes: g
h i uxl ¼ cos θxl xg cos θxl yg cos θxl zg
(2)
Fig. 1 Global (gx, gy, gz) and local (lx, ly, lz) frames. P represents a particle of the body of interest and gp and lp its position vectors in the two frames, respectively. Left superscripts denote the frame with respect to which a position vector or an orientation of a frame is represented. Subscripts denote the entity the vector or the matrix describes
20
V. Camomilla et al. g
h i uyl ¼ cos θyl xg cos θyl yg cos θyl zg
(3)
h i uzl ¼ cos θzl xg cos θzl yg cos θzl zg
(4)
g
Equations 2, 3, and 4 are normally represented in a matrix named orientation matrix (also rotation or attitude matrix): 2
cos θxl xg g Rl ¼ 4 cos θxl yg cos θxl zg
cos θyl xg cos θyl yg cos θyl zg
3 cos θzl xg cos θzl yg 5 cos θzl zg
(5)
It is useful to mention here that given the position vector and the orientation matrix of the local frame, relative to a global frame, and the position of a point (or of an ensemble of points, that is, a body) in the local frame (Fig. 1), it is possible to represent this position in the global frame through the following operator (coordinate transformation): g
p ¼ g Rl l p þ g ol
(6)
Of course the use of this operator may be generalized, since it allows representing the position of a point given in a reference frame to any other referred frame. It is important to note that the nine direction cosines are not independent; in fact the sum of the squares of each triplet equals one (the amplitude of the unit vector), and the vector product of each pair of unit vectors, being orthogonal, equals zero. The resulting six relationships leave only three of the nine direction cosines independent. In summary, the orientation of the local frame relative to the global frame is fully described by three scalars, which, added to the three scalars necessary to describe position (Eq. 1), makes a total of six scalars (the degrees of freedom of the body). The ensemble of position and orientation of a local frame and, thus, of a body, is named pose. Another and more compact way of representing the orientation of a body, i.e., of a local frame, is to exploit the Euler’s theorem and use the kinematic variables that describe the fictitious rotation that takes the local frame from being parallel to the global frame to the target orientation. We can achieve this by decomposing this rotation into three finite successive rotations about body-fixed axes in their current orientation. The corresponding three rotation angles, called Euler angles, completely describe the given rotation. There are twelve possible sets of Euler angles. Six imply the first and third rotation about the same axis (symmetric sets), and six imply rotations about the three different axes, in all possible combinations (asymmetric sets). The latter angle sets are also referred to as Cardan or Bryant angles. A second approach explicitly identifies the axis of rotation n (unit vector) and the angle of rotation θ that realize the target rotation and uses the following orientation (or rotation) vector: θ ¼ θn
(7)
Three-Dimensional Reconstruction of the Human Skeleton in Motion
21
Both the orientation vector and the Euler angles are illustrated in standard kinematics and dynamics texts and are not discussed here (Shuster 1993). However, a few remarks will be made in a subsequent section with regard to the use of these quantities to describe joint kinematics. After the above-illustrated representation of the instantaneous pose of a body, it is seen that, if the body can be hypothesized to keep its shape unaltered during the analyzed movement, the number of scalars necessary to describe it is reduced from 3 P N to 6 N + 3 P. Under these very favorable circumstances, we talk of a rigid body and use the part of classical mechanics that operates under the mentioned hypothesis. The question arises now whether the rigid body hypothesis is applicable to the constituents of the locomotor apparatus in motion. In this respect we have two orders of problems: one regards the mathematical modeling of the locomotor apparatus and the other the acquisition of experimental data. It is commonly accepted that if the focus of the investigation is skeletal motion, then bones may be considered to be rigid bodies without having a significant impact on the end results. When aiming at the estimation of internal loads or mechanical energy, then entire body segments are involved in the modeling exercise and may be disputable to consider them rigid bodies due to the fact that deformation and displacements occurring during movement of muscular and visceral masses may significantly change some inertia parameters with consequent inertial effects (mass moments of inertia and location of the center of mass; Clark and Hawkins 2010; Pain and Challis 2001; Zelik and Kuo 2010) (chapter ▶ “Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis”). It is evident that this issue becomes critical when high accelerations are involved (Challis and Pain 2008; Gruber et al. 1998; Liu and Nigg 2000; Riddick and Kuo 2016). An experimental issue consists on the fact that bone movement can be accurately recorded only using invasive techniques (intracortical pins or medical imaging involving the use of ionizing radiations). As better illustrated later, under normal circumstances, we can only capture the movement of 3-D markers (through stereophotogrammetry or magneto-inertial measurement units) attached to the skin surface above the bony segment of interest. When using stereophotogrammetry, the 3-D marker is constructed using three or more point markers (marker cluster). A 3-D marker is unavoidably mobile with respect to the underlying bone due to the interposed soft tissue, and, thus, its global motion defers from that of the bone. In general, this local displacement of the marker is caused by skin sliding associated joint movement, soft tissue volumetric deformation due to muscular contraction, gravity, and inertial effects on relevant masses (wobbling). In this context, the relative motion between marker and underlying bone is to be regarded as an artifact, the so-named soft tissue artifact, which, if not properly dealt with, has very serious consequences on the reliability of the results of the analysis (Cappozzo 1991; Garling et al. 2007; Lamberto et al. 2016; Leardini et al. 2005; Li et al. 2012). If the objective is a kinetic or energy analysis, then quantitative information concerning the abovementioned soft tissue relative motion, which is concerning the instantaneous body segment shape and mass distribution during movement and the estimate of the
22
V. Camomilla et al.
consequent time-varying inertia parameters, may be required posing an experimental problem very difficult to be dealt with using the present state of the art of measurement technology (Wakeling and Nigg 2001) (chapter ▶ “Simulation of Soft Tissue Loading From Observed Movement Dynamics”). In the general framework illustrated above and accepting the related limitations, in human and animal movement analysis, rigid body mechanics is normally used to model the locomotor apparatus. The objective of this chapter is to illustrate the way this is done. From now on we will refer to the estimate of bone pose being confident of the fact that, to all practical purposes, bones may be considered rigid bodies and that models of the relevant soft tissues may be associated to the bone in its current pose. Note that most analyses demand not only the estimate of the instantaneous pose of the bones involved but also the full reconstruction of their outer surface in the 3-D in silico space. It is important to acknowledge the fact that the quantitative description of human motion involves the use of kinematic and kinetic vector components, the values of which depend on the Cartesian reference axes with respect to which they are defined. Thus, for the sake of their repeatability, equally repeatable set of axes must be made available. As will be clearer later, it is easy to accomplish this requirement for the global frames, while, by cause of technical difficulties and morphological complexity, it is more difficult for the local frames associated with the bones (Della Croce et al. 1999, 2005). Furthermore, these sets of axes must be defined so that the resulting kinematic and kinetic scalar quantities have a clear functional meaning and effectively and consistently describe the anatomical and functional entities used in the medical-biological literature.
Global and Local Frames: Terminology As reported above, describing the skeletal-system movement involves the use of sets of coordinate system axes that define the global and the local frames. The definition and the methods for identifying these coordinate systems vary with the objective of the analysis and the measuring instrument involved.
Global Frames In a movement analysis laboratory, the following inertial, global frames can be defined (Fig. 2; Cappozzo et al. 1995, 1997b). Motion capture system frame: it is composed of the set of axes used by either the stereophotogrammetric system or the magneto-inertial system to represent pointmarker position or 3-D marker orientation, respectively. This is arbitrarily defined during the motion capture system calibration procedure.
Three-Dimensional Reconstruction of the Human Skeleton in Motion
23
Fig. 2 Movement analysis laboratory equipped with a stereophotogrammetric system and two force plates. Global frames are depicted. If a locomotor act is analyzed, the motor task frame may coincide with the frame of one of the two force plates
Motor task frame: this frame is consistent with the analyzed motor task and sometimes describes its basic features. For instance, when locomotor acts are investigated, one axis of the frame indicates the mean direction of progression. According to the general recommendations from the International Society of Biomechanics (Wu and Cavanagh 1995), in human locomotion analysis, right-handed orthogonal coordinate systems should have the x-axis pointing in the direction of progression, y pointing vertically upward, and z pointing to the right. Dynamometer frame: this is the frame in which force and moment components are given by the instrument and is defined during its calibration. Plumb line: this is a single axis and represents the orientation of the gravity line, usually assumed to point downward. Therefore, within the same experiment, different mechanical quantities can be measured with respect to different global frames. However, normally, their interpretation, or their use as input to musculoskeletal models allowing the estimation of further non-measurable quantities, requires that all of them be represented in the same frame (primary global frame). The latter role is usually assumed by the motor
24
V. Camomilla et al.
task frame. This procedure involves the determination of the position vector and the orientation matrix of all secondary global frames involved relative to the primary frame, allowing to obtain any vector quantity in the primary frame using Eq. 6. In order to achieve this result, the dynamometers and the pathway, staircase, or any other implement used to perform the motor task must be equipped with a 3-D marker, the pose of which may be detected by the motion capture system being used (Rabuffetti et al. 2003).
Local Frames We deal here with the frames associated with a bony segment (Fig. 3). Anatomical frame. This frame is made of orthogonal axes that exhibit the following properties: (i) compatibility with the anatomical axes and planes defined in the anatomy literature and (ii) intra- and inter-subject repeatability. The concept of repeatability must, of course, be extended to the portability of the results of movement analysis among different laboratories (Benedetti et al. 2013). This entails the standardization of the definition of the anatomical frames (Cappozzo et al. 1995; Wu and Cavanagh 1995; Wu et al. 2002). The following local frames depend on the specific technique used for the determination of their pose. For this reason they are named technical frames. Bone-embedded frame. This is a frame rigidly connected with the bone, but with no repeatable relationship with its anatomy. Motion technical frame. This is the frame the pose of which is provided by the motion capture system and related data processing. It, normally, represents an estimate of the bone-embedded frame. Fig. 3 Local frames
Three-Dimensional Reconstruction of the Human Skeleton in Motion
25
Morphology technical frame. This is the frame used by the measurement instrument (i.e., imaging apparatus, such as MRI or CT) that provides a digital model of the bone and, eventually, of the relevant soft tissues.
Estimate of the Bone-Embedded Frame Pose Bone movement may be directly tracked using medical imaging techniques. X-ray fluoroscopy, either planar or biplanar, is mostly used to record a single joint movement (Banks and Hodge 1996). This technique, however, exhibits the following limitations: it uses ionizing radiation, the measurement volume and the sampling frequency are limited, and the procedure is operator time-consuming. Recently systems have been developed that move alongside the subject, thus overcoming the limitation of the measurement volume, at least in one dimension (Guan et al. 2016) (chapter ▶ “Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography”). Other techniques are under development that exploit magnetic resonance and ultrasonic technologies (chapter ▶ “3D Musculoskeletal Kinematics Using Dynamic MRI”). In this chapter we deal only with the motion capture techniques mostly used in human movement analysis that track bone motion indirectly, but display the advantages of being noninvasive, having large measurement volumes and high sampling rates: stereophotogrammetry associated with skin markers, or magneto-inertial measurement units (Fig. 4). The fundamental concepts illustrated regarding the use of
Fig. 4 Input and output data of bone pose estimators using skin markers and stereophotogrammetry (a) or magneto-inertial measurement units (b)
26
V. Camomilla et al.
rigid body mechanics for the modeling of the human locomotor apparatus in motion are, however, valid whatever technique is used to monitor this motion.
Stereophotogrammetry Optoelectronic stereophotogrammetry is, to date, the most widely used solution for measuring skeletal kinematics. It is made of a number of video cameras, connected to a computer, the fields of view of which intersect defining the measurement volume. This system provides the position (Cartesian coordinates) of point markers, either emitting or retro-reflecting light, located on the skin surface relative to a global frame. This is done through mathematical operators that receive the 2-D coordinates of the point-marker images, measured in the image plane of at least two cameras at any given instant of time, and parameters that describe the location in space and optical features of the cameras. These parameters are obtained through the calibration of the stereophotogrammetric system. The reconstructed positions in the global frame of three or more nonaligned skin markers (marker cluster or 3-D marker) located above the bone of interest and a mathematical estimator are used to construct a motion technical frame and determine its pose relative to the global frame (Fig. 4) (chapters ▶ “3D Dynamic Pose Estimation from Markerless Optical Data” and ▶ “3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers”). Markers are located in such a way to comply with technical requirements such as visibility to a sufficient number of cameras, maximal relative distance between markers (Cappozzo et al. 1997a), and minimal relative movement between them and the underlying bone. Virtual markers may also be used to construct a marker cluster. A virtual marker is a point, the position of which is calculated using the positions of the available physical markers and a geometric or statistical rule. A typical example of virtual marker is a point approximating a joint center, which is in common between two adjacent bones. If the pose of one bony segment is successfully reconstructed, the global position time history of the virtual marker, the position of which in this segment is known, can be obtained using Eq. 6 and used to reconstruct the pose of the second body segment as if it were a physical marker. This procedure allows reducing the number of physical markers to be tracked. Note that, while this economical attitude could be justified years ago, this is not the case with current optical motion capture technology that can deal with a very high number of markers simultaneously. A marker cluster observed, for instance, in a selected instant of time or subject’s posture, is taken as a rigid cluster model. Using some geometric rule, a set of orthogonal axes is associated to it (model frame). If the skin markers were rigidly attached to the underlying bone, the rigid cluster model could be superimposed onto the current marker cluster and the model frame taken as motion technical frame. Unfortunately, this is not the case: during movement, the reconstructed
Three-Dimensional Reconstruction of the Human Skeleton in Motion
27
Fig. 5 The artifact displacements of the skin markers (arrows), occurring in a given interval of time, result in a rigid transformation (a translation plus a rotation) plus a nonrigid transformation (a change in size and shape) of the marker cluster (Andersen et al. 2012; Barré et al. 2013; Benoit et al. 2015; De Rosario et al. 2012; Dumas et al. 2014; Grimpampi et al. 2014). The former transformation has been proved to be the dominant part (Andersen et al. 2012; Barré et al. 2013; Benoit et al. 2015; Bonci et al. 2015; Dumas et al. 2015)
marker local positions undergo variations due to stereophotogrammetric errors (Chiari et al. 2005) and, as already mentioned, the soft tissue deformation (Leardini et al. 2005). Thus, the marker cluster changes its shape and moves relative to the bone (Fig. 5). As a consequence, the above-mentioned superimposition exercise must be carried out using an approximation approach (“single-body optimization”). This consists in a least-squares method, the most popular of which minimizes the sum of the squared distances between corresponding points of the cluster model and of the marker cluster, the so-named Procrustes distance (Soderkvist and Wedin 1993). In this way the instantaneous pose of the motion technical frame can be determined and taken as an estimate of the bone-embedded frame. Instead of dealing with one marker cluster, and thus one bone, at a time, it is possible to carry out the least-squares superimposition procedure illustrated above using a multi-body system made of a chain of rigid cluster models connected by kinematic models of the joints involved (“multi-body kinematics optimization”). The joint models proposed in the literature typically involve major simplifications with respect to real and subject-specific joints and have less than six independent degrees of freedom (Charlton et al. 2004; Duprey et al. 2010; Li et al. 2012; Lu and O’Connor 1999; Reinbolt et al. 2005; Richard et al. 2016; Scheys et al. 2011). It should be emphasized that while instrumental errors that are random in nature have no significant effect on the end result, the illustrated procedures (single-body and multi-body kinematics optimization) do not attenuate the propagation of the largest portion of the soft tissue artifact (Andersen et al. 2010; Bonci et al. 2015; Clément et al. 2015, 2017; Gasparutto et al. 2015; Li et al. 2012). The latter issue remains the greatest obstacle to the accurate reconstruction of skeletal movement and deserves further attention by the human movement analyst community.
28
V. Camomilla et al.
Magneto-Inertial Measurement Units Miniature magneto-inertial measurement units, embedding a microprocessor and often endowed with wireless communication technology, are an increasingly popular alternative to stereophotogrammetry for 3-D human movement analysis (chapter ▶ “Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units”). These units are attached to the body segment of interest. They comprise three-axis linear accelerometer and angular rate sensors and a three-axis magnetometer. The physical quantities provided by each sensor are measured with respect to the axes of a unit-embedded frame generally aligned with the edges of the unit case. Through algorithms able to fuse the redundant information available and compensate for sensor noise and drift, the 3-D orientation of the unit-embedded frame relative to a global frame is provided. The unit-embedded frame is assumed to be the motion technical frame and, again, despite the soft tissue artifact, is assumed to be an estimate of the bone-embedded frame (Fig. 4). As opposed to stereophotogrammetry, magneto-inertial measurement units do not supply reliable positional information. The advantage of this technique is that it does not impose limits to the measurement volume.
Anatomical Calibration So far we have seen how a bone-embedded frame, as observed during the execution of a motor task, may be reconstructed in silico in its current pose. The next step is the representation of a digital model of the relevant bone in this frame. In order to comply with the repeatability issues discussed in a previous section, this digital model must carry an anatomical frame with it. The experimental acquisition of the parameters that allow the construction in silico of the bone model in its current pose and of the relevant anatomical axes is referred to as “anatomical calibration.”
Subject-Specific Morphology Data The 3-D digital model of the bone of interest may be defined at different levels of resolution that depend on the number of points used to describe the bone external surface. The minimum number of these points is 3, as imposed by the possibility of constructing an anatomical frame. A larger number of points allow a more realistic rendering of the bone in silico and a more accurate association of the soft tissue digital models to the bone. A subject-specific bone digital model may be measured or estimated using medical imaging. A full reconstruction of that model may be obtained using magnetic resonance, but this is rarely possible for logistic and economic reasons. A way of estimating a subject-specific bone digital model, with sufficient accuracy for most purposes, is through two planar X-rays of the bone and a relevant statistical model and shape recognition algorithm (Chaibi et al. 2012). This method is made applicable
Three-Dimensional Reconstruction of the Human Skeleton in Motion
29
by a low-dose X-ray imaging technology performed at a low dose and with an expanded dynamic range that allows for whole-body scanning while the subject is under normal weight-bearing conditions (Melhem et al. 2016). It is evident that with all methods that involve medical imaging, the bone digital model is represented in a morphology technical frame that is different from the motion technical frame. For a number of reasons that will be illustrated later, the subject-specific bone digital model must carry labeled anatomical landmarks (sites with recognizable anatomical features). Given a clear definition of these landmarks, this labeling may be carried out in silico through a virtual palpation (Van Sint Jan et al. 2003). An alternative approach is the estimate of the subject-specific bone model using subject-specific partial information collected in the movement analysis laboratory and a template bone model that is made to match the above-mentioned morphological information. This partial information may consist of the 3-D position of isolated points or clouds of adjacent points that lie on the bone surface and of lines oriented as anatomical axes or lie in anatomical planes. In this case, morphological information is available in the motion technical frames. The position of anatomical landmarks may be determined with an ad hoc stereophotogrammetric acquisition by either temporarily locating skin markers over them (Cappozzo 1984) or using a wand equipped with a marker cluster (Fig. 6a; Cappozzo et al. 1995), after identification by manual palpation (Van Sint Jan et al. 2003). In some experimental protocols, skin
Fig. 6 Anatomical calibration in the stereophotogrammetric laboratory. (a) Skin markers are located consistently with technical requirements; anatomical landmarks are identified by manual palpation and their position measured using a wand. (b) Some skin markers are located on anatomical landmarks or lying in an anatomical plane. (c) As in (a), but portions of the bone surfaces are digitized by moving the tip of the wand over them. The center of the acetabulum is estimated using mathematical models or identified through a functional approach
30
V. Camomilla et al.
markers used to track motion are located over anatomical landmarks, the position of which is therefore readily available, or in an anatomical plane (Fig. 6b; Davis et al. 1991; Frigo et al. 1998; Leardini et al. 2007) (chapters ▶ “The Conventional Gait Model - Success and Limitations” and ▶ “Variations of Marker Sets and Models for Standard Gait Analysis”). More detailed morphological information may be acquired by using the above-mentioned wand over clouds of points of the bone covered with a layer of soft tissue that allows for their palpation through the skin (Fig. 6c; Donati et al. 2007, 2008). To these points, internal anatomical landmarks may be added when noninvasively identifiable. The position of these landmarks may be determined as a function of the position of other accessible anatomical landmarks and/or readily available anthropometric measures using population statistical models. Typical examples in this respect are the regression equations used to determine the position of the center of the acetabulum (Bell et al. 1990; Davis et al. 1991; Hara et al. 2016; Harrington et al. 2007; Leardini et al. 1999; Seidel et al. 1995) or the position of the clavicle, scapula, and humeral bone (Sholukha et al. 2009). In some cases, an internal anatomical landmark may be considered to coincide with a joint center of rotation (▶ “NextGeneration Models Using Optimized Joint Center Location”). This allows determining the position of these landmarks using a “functional approach.” This entails performing an ad hoc experiment during which the subject, equipped with suitable markers (skin-marker clusters or magneto-inertial measurement units), is asked to execute a 3-D movement of the joint involved. The time history of the pose of one bone relative to the other forming the joint is estimated and used to determine the position of the joint center and, thus, of the relevant anatomical landmark (Crabolu et al. 2016; Halvorsen 2003; McGinnis and Perkins 2013). An example in this respect is the center of the acetabulum that can be assumed to coincide with the center of the femoral head and the center of rotation of the femur relative to the pelvic bone (Fig. 7; Camomilla et al. 2006; Cappozzo 1984; Cereatti et al. 2010; Kainz et al. 2015; Leardini et al. 1999; Piazza et al. 2004). Similarly, the center of the Fig. 7 Movement normally used for the estimate of the center of the acetabulum using the functional approach. The figure depicts the trajectory of a point located on the distal portion of the femur and represented in the transverse plane of the pelvis. Numbers represent the movement sequence (Camomilla et al. 2006)
Three-Dimensional Reconstruction of the Human Skeleton in Motion
31
head of the humerus may be identified as the rotation center of the glenohumeral joint (Campbell et al. 2009; Lempereur et al. 2010). Lines that lie in anatomical planes also carry useful morphological information. By suitably choosing a joint movement, the resulting rotation axis may be assumed to have this property. Such an axis may be determined through the already mentioned functional approach (Gamage and Lasenby 2002; Halvorsen et al. 1999). Examples in this respect are the knee (Colle et al. 2016; De Rosario et al. 2017; Ehrig et al. 2007), the elbow joint (Fraysse and Thewlis 2014), and the talocrural joint (Sheehan 2010; van den Bogert et al. 1994) moving in the sagittal plane. The estimated rotation axis is assumed to lie in the frontal plane and, together with at least an anatomical landmark, allows the determination of this plane. It is worth emphasizing that, as opposed to what in some cases may appear in the literature, an axis or line, obtained using this method, is not an anatomical axis per se, although sometime it may be considered to approximate it. When magneto-inertial measurement units are used, the position of anatomical landmarks cannot be reliably measured. The only morphological information that can be collected concerns the orientation of lines that lie in anatomical planes. This information can be obtained using either a functional approach (Bouvier et al. 2015; Cutti et al. 2008, 2010; Favre et al. 2009; Luinge et al. 2007; Seel et al. 2012) or a specifically designed calibration device consisting in a rod carrying a magneto-inertial measurement unit and two mobile pointers perpendicular to it (Fig. 8). This unit provides the orientation relative to the motion technical frame, made available by the unit mounted on the body segment of interest, of a line joining two palpable anatomical landmarks pointed by the calibration device (Picerno et al. 2008).
Movement-Morphology Data Registration After the subject-specific morphological information is collected, the entire subjectspecific digital model of the bone must be associated to the motion technical frame in each sampled instant of time during the analyzed movement and, therefore, represented in silico in its current pose relative to the global frame. When the entire subject-specific digital model is made available, either measured or estimated through medical imaging, and represented in a morphology technical frame different from the motion technical frame, a transformation of the position vectors given in the former frame into position vectors in the latter frame must be carried out (the procedure is known as movement-morphology data registration). To this purpose Eq. 6 can be used, provided that the anatomical calibration procedure includes the determination of the orientation matrix and origin position vector of one frame with respect to the other. This may be achieved by having the position of at least three points in both frames and applying a superimposition procedure similar to that described in section “Estimate of the Bone-Embedded Frame Pose.” These points may be anatomical landmarks or, when using stereophotogrammetry, the skin markers. In the latter circumstance, the medical imaging procedure must be
32
V. Camomilla et al.
Fig. 8 Anatomical calibration using magneto-inertial measurement units. The orientations of the two dashed lines shown in the figure are detected by the unit mounted on the calibration device and define the sagittal plane of the femur. The line joining the medial epicondyle (LE) and the greater trochanter (GT) may be used for the orientation of the y-axis. The orientation of the other two axes is, thus, available as well. If the center of the femoral head is identified using the functional approach (Crabolu et al. 2016), a digital model of the femur may be matched with the available subject-specific morphological information and the definition of the femoral anatomical frame illustrated in Fig. 9 may be used
carried out with the subject carrying the marker set that will be used for the motion tracking. While doing this with the subject lying on a horizontal surface, as normally occurs with magnetic resonance, we may incur in a problem. His/her soft tissues are deformed in a different fashion then when assuming the static or dynamic postures under analysis. The thus obtained pose of the motion technical frame relative to the morphological technical frame is, therefore, different from that occurring during the motor task of interest. If partial subject-specific morphology information is collected in the motion capture laboratory, then the subject-specific bone model in its current pose may be obtained by matching a suitably chosen template bone model, provided in a morphology technical frame, with the available morphological data available in the motion technical frame. To this purpose a superimposition procedure associated with non-isomorphic scaling and a reorientation of the template model may be used (Chaibi et al. 2012; Donati et al. 2007, 2008; Quijano et al. 2013). In this case, the bone model is already represented in the motion technical frame, and, therefore, no registration procedure is required.
Three-Dimensional Reconstruction of the Human Skeleton in Motion
33
The bone model thus obtained normally carries a number of labeled anatomical landmarks or lines, those that have been used for its construction and registration. If the position of all anatomical landmarks required for further processing is not provided by this procedure, then a virtual palpation or geometric construction is carried out on the bone model. An example of geometrical construction is that of a line that may be supposed to lie in the frontal plane of the femur. This is the so-named cylinder axis, which is the axis that joins the centers of the two spheres that fit the medial and lateral posterior condyles of the femur and that is supposed to lie in the femur frontal plane (Yin et al. 2015).
Construction of the Anatomical Frame Once the subject-specific digital model of a bone is available, the relevant anatomical frame must be constructed. As already mentioned, this is done using the position of anatomical landmarks, and/or points or lines that lie on an anatomical plane, and a geometric rule. The anatomical frame may be constructed in the morphology technical frame or in the motion technical frame and, then, represented in whatever other frame it is required through the transformation represented by Eq. 6. It is evident that, given the many anatomical landmarks and other morphological features of the bone that can be made available, many anatomical sets of axes may be defined and have, in fact, been defined and illustrated in the literature (Fig. 9; Cappozzo et al. 1995; Kadaba et al. 1990; Wu and Cavanagh 1995; Wu et al. 2002). As already mentioned, while defining the term “anatomical frame,” the issue here is the inter- and intra-subject repeatability of this set of axes and their portability. Critical factors are the following. Anatomical landmarks are areas and not points, as assumed to be in the present construction, and their definition may slightly change depending on the source that describes them. When an operator palpates an external anatomical landmark, these circumstances may lead to large inter- and intra-operator variability as quantified in Della Croce et al. (1999). The uncertainty that affects internal anatomical landmarks or lines, such as those assumed to coincide with a joint center or a joint rotation axis, may also impact very negatively on repeatability (Stagni et al. 2000). By merging the human movement and the ever-developing medical imaging laboratories, we may be able, in the next future, to find better solutions to this problem. In addition, these enhanced solutions should be shared within the human movement analyst community and possibly undergo a standardization process for the sake of data portability.
Joint Kinematics Estimation The assessment of the stability and mobility of a joint is based on the observation of the relative movement of the two adjacent bones involved, that is, of joint kinematics. For each bone of interest, the procedures described in the previous sections
34
V. Camomilla et al.
Fig. 9 Example of anatomical frame definition (Cappozzo et al. 1995). Right-handed set of axes with the following characteristics (only two axes are defined because the third axis is the vector product of them). (a) Pelvis; origin: midpoint between the right anterior superior iliac spine (RASIS) and left anterior superior iliac spine (LASIS); zp-axis: oriented as the line passing through the RASIS and the LASIS with positive direction to the right; xp-axis: lies in the plane defined by the RASIS and LASIS and the midpoint between RPSIS and LPSIS and with its positive direction forward. (b) Femur; origin: midpoint between the lateral epicondyle (LE) and medial epicondyle (ME); yf-axis: it joins the origin with the center of the femoral head (FH) and its positive direction is proximal; zf-axis: it lies in the plane defined by FH, ME, and LE, and its positive direction is from left to right. (c) Tibia and fibula; origin: midpoint between the lateral malleolus (LM) and medial malleolus (MM); ys-axis: it is the line of intersection between the plane defined by the head of the fibula (HF), LM and MM (frontal plane), and the plane orthogonal to it passing through the tibial tuberosity (TT), LM, and MM; its positive direction is proximal; zs-axis: it lies in the frontal plane with positive direction from left to right
provide an estimate of the pose of an anatomical frame (only orientation if magneto-inertial sensors are used) relative to a global frame of choice in each sampled instant of time during movement (Eqs. 1 and 5). As already mentioned in a previous section, the relative pose between two rigid bodies is described by six scalar quantities, the six degrees of freedom; three describing the mutual orientation and three the mutual position. The objective of this section is to review the possible descriptions of joint kinematics and the issues associated with them. Given the orientation matrices and the position vectors of the anatomical frames of the proximal and distal bones of the joint under analysis relative to a global frame in any given instant of time (Fig. 10) g
Rp g op and g Rd g od ,
(8)
Three-Dimensional Reconstruction of the Human Skeleton in Motion
35
Fig. 10 Proximal and distal bone anatomical frames, their pose relative to the global frame (gRp, gop, and gRd, god respectively) and the pose of the distal bone relative to the proximal bone (pRd, pod)
we may represent the position vector of any given point in the global frame as a function of its position vector expressed in both local frames: p ¼ g Rp p p þ g op
(9)
p ¼ g Rd d p þ g od:
(10)
g
and g
By eliminating gp in Eqs. 9 and 10 and pre-multiplying both left and right terms of the resulting equation by the transposed matrix g RpT , p p may be calculated as p
p ¼ g RpT g Rd d p þ g RpT g od g op :
(11)
From this equation, and keeping in mind Eq. 6, it results that the orientation and position of the distal bone relative to the proximal bone, are given by p
Rd ¼ g RpT g Rd and p od ¼ g RpT g od g op , respectively:
(12)
Although pRd and pod fully describe joint kinematics, the scalar quantities that appear in them do not necessarily comply with the requirements of consistency with the anatomical and physiological terminology and thus of an effective description of function. This issue is tackled as illustrated in the following.
36
V. Camomilla et al.
Translational Degrees of Freedom The relative position of two adjacent bones in a given instant of time is described making reference to a vector (t) joining the position of a point defined in the proximal (Kp) and that of a point defined in the distal (Kd) local frames. These two points are normally chosen so that they coincide while the subject assumes a reference posture (zero joint translation). The translational degrees of freedom are described by the way the three components of vector t vary in time during the movement. Since during function translation is coupled with rotation, vector t depends on the location of the above-mentioned points. This location must therefore be chosen with care and accounting for the characteristics of the joint dealt with and based on the objective of the analysis. For instance, when dealing with the hip joint, Kd and Kp may be defined as coinciding with the mean center of rotation of that joint and associated with both pelvic bone and femur in a selected instant of time during movement or during a selected posture (Cereatti et al. 2010). If the analysis regards a joint that is more complex from the mechanical point of view, such as the tibiofemoral joint, than the midpoint of the transepicondylar axis (Fig. 11; Grood and Suntay 1983), a contact point between the two bones involved may be of interest. The other choice that needs to be made regards the set of axes with respect to which vector t is represented, again in each sampled instant of time. This set of axes may be that of the proximal or distal anatomical frame. Normally, the axes used to describe the three rotational degrees of freedom, illustrated later, are used in their current orientation. It is worthwhile noticing that, due to the fact that joint linear displacements are normally smaller or in the order of the resolution of most motion capture systems, there is little literature dealing with them and no shared convention for their description.
Rotational Degrees of Freedom The quantification of the rotational degrees of freedom is typically based on the observation, in the sampled instants of time during movement, of the orientation of Fig. 11 A possible definition of knee translation vector (t) during flexion. Kp is made to coincide with the midpoint of the transepicondylar axis; Kd and Kp are rigid with the tibia and with the femur, respectively, and they coincide during a selected reference posture
Three-Dimensional Reconstruction of the Human Skeleton in Motion
37
the distal bone relative to the proximal bone using a sequence of three rotations about the distal bone anatomical axes. The rotation sequence proposed by Grood and Suntay (1983) and recommended by Wu and Cavanagh (1995) and Wu et al. (2002), that is, the Cardan sequence, yields three angles that best match the way that functional anatomy uses for most joints (Fig. 12): (i) angle γ around zd (coinciding with zp) for flexion-extension, (ii) angle α around the current orientation of xd for abduction-adduction, and (iii) angle β around the current orientation of yd for internal-external rotation. If, during the illustrated sequence, each of these three axes is frozen with the orientation it assumes during the related rotation, a set of non-orthogonal axes is obtained (Fig. 12). The axis (xd) around which the second rotation occurs is referred to as floating axis, and the three axes are named joint axes. The equations that allow the calculation of the Cardan angles from the orientation matrix pRd are α ¼ sin 1p Rd ð3, 2Þ β ¼ sin 1 ½p Rd ð3, 1Þ= cos α
(13)
γ ¼ sin 1 ½p Rd ð1, 2Þ= cos α The demonstration of these equations can be found in the standard mechanical literature. Note that, if the second rotation equals π/2 or a multiple of it, then we incur in a singularity condition (gimbal lock). This circumstance prevents the use of the Cardan convention for those joints which, during movement, may undergo an abduction-adduction greater than π/2.
Fig. 12 The Cardan angles that describe an orientation of the distal bone relative to the proximal bone are defined according to the following sequence of operations: (i) to begin with the anatomical frames of the proximal ( p, hip bone) and distal bone (d, femur) are parallel; (ii) the first rotation γ occurs around the dz-axis; (iii) the second rotation α around the dx-axis in the orientation, it assumes after the first rotation (d1x); (iv) the third rotation β around the dy-axis in the orientation, it assumes after the second rotation (d2y). Axes in red (dz, d1x, d2y) are the joint axes and d1x is the floating axis
38
V. Camomilla et al.
Moreover, using this method, the instantaneous orientation of the distal bone is represented starting from a reference orientation that is equal to that of the proximal bone. Since during upright posture, the two anatomical frames are not necessarily parallel; the corresponding joint angles are not zero as often assumed to be in the biomedical literature. It is also important to remark that the three angles thus obtained univocally describe an orientation, but not true rotations. In addition, the Cardan angles, as more in general the Euler angles, do not enjoy the additive property. This means that, in principle, we cannot subtract two values of a given Cardan angle, observed in two different instants of time during movement or different postures, and, indeed, the result cannot be interpreted as an angular displacement, nor can we calculate relevant arithmetical averages (Pierrynowski and Ball 2009). For the same reason, when plotting a Cardan angle versus time, a continuous line should not be made to join two sampled values because this would imply an interpolation between them. We may conclude that, paraphrasing what Churchill is told to have said with regard to democracy, the Cardan angles is the worst form of joint rotation description, except for all the others. The second approach, definitely less popular than and as abstract as the previous one, is based on the orientation vector pθd applied to the orientation, observed in the sampled instants of time during movement, of the distal bone relative to the proximal bone anatomical frame (Fioretti et al. 1997; Woltring 1994). This vector may be represented in any set of local axes of choice, the proximal bone or the distal bone anatomical axes, or the non-orthogonal joint axes. This choice shall be tailored upon the consistency of the three resulting angles with functional anatomy. A positive property of this method is that it does not suffer the gimbal lock issue, but displays all other limitations of the Cardan angles. A third method to describe joint kinematics uses the following geometric approach (Paul 1992). The flexion-extension rotation is measured through the angle formed by the y-axis of the distal bone and the projection onto the xy (sagittal) plane of this bone of the y-axis of the proximal bone. The abduction-adduction rotation is given by the angle formed by the y-axis of the distal bone and the projection of the y-axis of the proximal bone onto the yz (frontal) plane of the distal bone. The internal-external rotation angle is the angle formed by the x-axis of the distal bone and the projection of the x-axis of the proximal bone onto the xz (transverse) plane of the distal bone. This approach is very intuitive; in fact it is similar to the way used in functional anatomy to describe joint motion. It is as abstract as the other methods, but it may deserve some attention from the human movement analyst community.
Precision and Accuracy of Joint Kinematics Estimates Two major issues affect the estimate of joint kinematics. One impacts on accuracy and it is mainly associated with the soft tissue artifact. The other affects precision and depends on the variability with which anatomical landmarks and, therefore, anatomical frames are determined.
Three-Dimensional Reconstruction of the Human Skeleton in Motion
39
Fig. 13 The six degrees of freedom of the knee joint during the support phase of running of a lean young adult able-bodied male subject. Black lines represent reference values obtained using clusters of markers mounted on intracortical pins (Cereatti et al. 2017). Gray lines represent the values obtained using clusters of skin markers and a single-body least-squares bone pose estimator (Camomilla et al. 2015)
In a preceding section, we have illustrated the origin of the soft tissue artifact and how and in what measure it impacts the estimate of bone pose while using noninvasive motion capture techniques. Figure 13 shows an example of the propagation of this artifact to the six degrees of freedom of the knee during the support phase of running, when the single-body optimization method previously illustrated is used. Inaccuracies in the order of 10 and 10 mm for rotations and linear displacements, respectively, may be observed. As far as the precision issue is concerned, in Table 1 we report data that provide an idea of the impact that the variability in the identification of the orientation of the mediolateral anatomical axis of the femur may have on knee kinematics during level walking as described using for different conventions.
Future Directions Although the potential of quantitative movement analysis in research, professional decision-making, and intervention practice is fully recognized, its application to large portions of the locomotor apparatus that move in sizeable volumes is currently
40
V. Camomilla et al.
Table 1 Maximum variation of the knee joint angles (degrees) with respect to the reference values as a result of the variation of the orientation of the mediolateral anatomical axis of the femur over the range 15 (Fioretti et al. 1997). Method used for the joint kinematics description: (1) Cardan angles, (2) projection of the joint orientation vector onto the femoral anatomical axes, (3) projection of the joint orientation vector onto the knee joint axes, and (4) geometric approach (see previous section for relevant details) Method 1 2 3 4
Flexion-extension 1.0 3.9 2.5 1.6
Abduction-adduction 13.8 17.5 12.4 14.3
Internal-external rotation 9.0 0.0 6.5 21.0
limited by several problems. The experimental and analytical protocols, which have been described in this chapter, provide results regarding skeletal motion with precision and accuracy that are insufficient to answer many of the questions posed by scientists and professionals. At the moment, however, using optical or magnetoinertial technology, the reconstruction of skeletal movement may be carried out with errors that may be in the order of 10 mm and 10 or more, for position and orientation, respectively. This collides with the fact that some relative movements between adjacent joints, among those of great interest in several applications, are in the order of a few millimeters and degrees. Thus, future research must aim at this resolution by exploiting emerging technology (including other domains, e.g., medical imaging, robotics, animation) and innovative experimental and analytical methods. In particular it should focus on the following issues: (a) Development of algorithms able to process the data provided by the motion capture systems and make data describing the movement of the skeleton with the mentioned accuracy available. A specific challenge is the compensation for the artifact movement between the superficially tracked motion and the underlying structures. Two approaches are being attempted. One uses “intelligent markers” which, through an ultrasound technology, gathers trial-specific information about this artifact movement (Masum et al. 2016). Another approach uses bone pose estimators, either single-body or multi-body kinematics optimization, that embed a mathematical model of the artifact (Bonnet et al. 2017; Camomilla et al. 2015). (b) Devising non- or minimally invasive experimental techniques, possibly usable in the movement analysis laboratory, and statistical shape analysis able to provide accurate and detailed subject-specific bone and soft tissue morphology to be registered with motion data. (c) The definition and construction of anatomical frames that allow for an adequate repeatability of the results of the analysis and their standardization for the sake of data portability. (d) Implementation of advanced probabilistic approaches that accommodate for intraindividual variability and input data uncertainty.
Three-Dimensional Reconstruction of the Human Skeleton in Motion
41
In addition, other open issues are the following. Computational models of the neuromusculoskeletal system are not as integrative across space scales as demanded by a holistic approach and encounter difficulties in incorporating the characteristics of a specific subject. Finally, there is no consensus on the optimal metrics to use when assessing motor function.
References Andersen MS, Benoit DL, Damsgaard M, Ramsey DK, Rasmussen J (2010) Do kinematic models reduce the effects of soft tissue artefacts in skin marker-based motion analysis? An in vivo study of knee kinematics. J Biomech 43:268–273 Andersen MS, Damsgaard M, Rasmussen J, Ramsey DK, Benoit DL (2012) A linear soft tissue artefact model for human movement analysis: proof of concept using in vivo data. Gait Posture 35:606–611 Banks SA, Hodge WA (1996) Accurate measurement of three-dimensional knee replacement kinematics using single-plane fluoroscopy. IEEE Trans Biomed Eng 43:638–649 Barré A, Thiran JP, Jolles BM, Theumann N, Aminian K (2013) Soft tissue artifact assessment during treadmill walking in subjects with total knee arthroplasty. IEEE Trans Biomed Eng 60:3131–3140 Bell AL, Pedersen DR, Brand RA (1990) A comparison of the accuracy of several hip joint center location prediction methods. J Biomech 23:617–621 Benedetti MG, Merlo A, Leardini A (2013) Inter-laboratory consistency of gait analysis measurements. Gait Posture 38:934–939 Benoit DL, Damsgaard M, Andersen MS (2015) Surface marker cluster translation, rotation, scaling and deformation: their contribution to soft tissue artefact and impact on knee joint kinematics. J Biomech 48:2124–2129 van den Bogert AJ, Smith GD, Nigg BM (1994) In vivo determination of the anatomical axes of the ankle joint complex: an optimization approach. J Biomech 27:1477–1488 Bonci T, Camomilla V, Dumas R, Chèze L, Cappozzo A (2015) Rigid and non-rigid geometrical transformations of a marker-cluster and their impact on bone-pose estimation. J Biomech 48:4166–4172 Bonnet V, Richard V, Camomilla V, Venture G, Cappozzo A, Dumas R (2017) Joint kinematics estimation using a multi-body kinematics optimisation and an extended Kalman filter, and embedding a soft tissue artefact model. J Biomech (in press) Bouvier B, Duprey S, Claudon L, Dumas R, Savescu A (2015) Upper limb kinematics using inertial and magnetic sensors: comparison of sensor-to-segment calibrations. Sensors 15:18813–18833 Camomilla V, Cereatti A, Vannozzi G, Cappozzo A (2006) An optimized protocol for hip joint centre determination using the functional method. J Biomech 39:1096–1106 Camomilla V, Bonci T, Dumas R, Chèze L, Cappozzo A (2015) A model of the soft tissue artefact rigid component. J Biomech 48:1752–1759 Campbell AC, Lloyd DG, Alderson JA, Elliott BC (2009) MRI development and validation of two new predictive methods of glenohumeral joint center location identification and comparison with established techniques. J Biomech 42:1527–1532 Cappozzo A (1984) Gait analysis methodology. Hum Mov Sci 3:27–50 Cappozzo A (1991) Three-dimensional analysis of human locomotor acts: experimental methods and associated. Hum Mov Sci 10:589–602 Cappozzo A, Catani F, Della Croce U, Leardini A (1995) Position and orientation in space of bones during movement, anatomical frame definition and determination. Clin Biomech 10:171–178 Cappozzo A, Cappello A, Della Croce U, Pensalfini F (1997a) Surface-marker cluster design criteria for 3-D bone movement reconstruction. IEEE Trans Biomed Eng 44:1165–1174
42
V. Camomilla et al.
Cappozzo A, Della Croce U, Lucchetti L (1997b) Gait data, terminology and definition. In: Allard P, Cappozzo A, Lumberg A, Vaughan K (eds) Three-dimensional analysis of human locomotion. Wiley, New York, pp 129–132 Cereatti A, Margheritini F, Donati M, Cappozzo A (2010) Is the human acetabulofemoral joint spherical? J Bone Joint Surg (Br) 92:311–314 Cereatti A, Bonci T, Akbarshahi M, Aminian K, Barré A, Begon M, Benoit DL, Charbonnier C, Dal Maso F, Fantozzi S, Lin CC, Lu TW, Pandy MG, Stagni R, van den Bogert AJ, Camomilla V (2017) Standardization proposal of soft tissue artefact description for data sharing in human motion measurements. J Biomech. https://doi.org/10.1016/j.jbiomech.2017.02.004 Chaibi Y, Cresson T, Aubert B, Hausselle J, Neyret P, Hauger O, de Guise JA, Skalli W (2012) Fast 3D reconstruction of the lower limb using a parametric model and statistical inferences and clinical measurements calculation from biplanar X-rays. Comput Methods Biomech Biomed Eng 15:457–466 Challis JH, Pain MTG (2008) Soft tissue motion influences skeletal loads during impacts. Exerc Sport Sci Rev 36:71–75 Charlton IW, Tate P, Smyth P, Roren L (2004) Repeatability of an optimised lower body model. Gait Posture 20:213–221 Chiari L, Cappozzo A, Della Croce U, Leardini A (2005) Human movement analysis using stereophotogrammetry. Part 2: experimental errors. Gait Posture 21:197–211 Clark T, Hawkins D (2010) Are fixed limb inertial models valid for dynamic simulations of human movement? J Biomech 43:2695–2701 Clément J, Dumas R, Hagemesiter N, de Guise JA (2015) Soft tissue artifact compensation in knee kinematics by multi-body optimization: performance of subject-specific knee joint models. J Biomech 48:3796–3802 Clément J, Dumas R, Hagemeister N, de Guise JA (2017) Can generic knee joint models improve the measurement of osteoarthritic knee kinematics during squatting activity? Comput Methods Biomech Biomed Eng 20:94–103 Colle F, Lopomo N, Visani A, Zaffagnini S, Marcacci M (2016) Comparison of three formal methods used to estimate the functional axis of rotation: an extensive in-vivo analysis performed on the knee joint. Comput Methods Biomech Biomed Eng 19:484–492 Crabolu M, Pani D, Raffo L, Cereatti A (2016) Estimation of the center of rotation using wearable magneto-inertial sensors. J Biomech 16:3928–3933 Cutti AG, Giovanardi A, Rocchi L, Davalli A, Sacchetti R (2008) Ambulatory measurement of shoulder and elbow kinematics through inertial and magnetic sensors. Med Biol Eng Comput 46:169–178 Cutti AG, Ferrari A, Garofalo P, Raggi M, Cappello A, Ferrari A (2010) “Outwalk”: a protocol for clinical gait analysis based on inertial and magnetic sensors. Med Biol Eng Comput 48:17–25 Davis RB, Ounpuu S, Tyburskky D, Gage R (1991) A gait analysis data collection and reduction technique. Hum Mov Sci 10:575–587 De Rosario H, Page A, Besa A, Mata V, Conejero E (2012) Kinematic description of soft tissue artifacts: quantifying rigid versus deformation components and their relation with bone motion. Med Biol Eng Comput 50:1173–1181 De Rosario H, Page A, Besa A (2017) Analytical study of the effects of soft tissue artefacts on functional techniques to define axes of rotation. J Biomech. https://doi.org/10.1016/j. jbiomech.2017.01.046 Della Croce U, Cappozzo A, Kerrigan DC (1999) Pelvis and lower limb anatomical landmark calibration precision and its propagation to bone geometry and joint kinematics. Med Biol Eng Comput 37:155–161 Della Croce U, Leardini A, Chiari L, Cappozzo A (2005) Human movement analysis using stereophotogrammetry. Part 4: assessment of anatomical landmark mislocation and its effects on joint kinematics. Gait Posture 21:226–237
Three-Dimensional Reconstruction of the Human Skeleton in Motion
43
Donati M, Camomilla V, Vannozzi G, Cappozzo A (2007) Enhanced anatomical calibration in human movement analysis. Gait Posture 26:179–185 Donati M, Camomilla V, Vannozzi G, Cappozzo A (2008) Anatomical frame identification and reconstruction for repeatable lower limb joint kinematics estimates. J Biomech 41:2219–2226 Dumas R, Camomilla V, Bonci T, Chèze L, Cappozzo A (2014) Generalized mathematical representation of the soft tissue artefact. J Biomech 47:476–481 Dumas R, Camomilla V, Bonci T, Chèze L, Cappozzo A (2015) What portion of the soft tissue artefact requires compensation when estimating joint kinematics? J Biomech Eng 137:064502. https://doi.org/10.1115/1.4030363 Duprey S, Chèze L, Dumas R (2010) Influence of joint constraints on lower limb kinematics estimation from skin markers using global optimization. J Biomech 43:2858–2862 Ehrig RM, Taylor WR, Duda GN, Heller MO (2007) A survey of formal methods for determining functional joint axes. J Biomech 40:2150–2157 Favre J, Aissaoui R, Jolles BM, de Guise JA, Aminian K (2009) Functional calibration procedure for 3D knee joint angle description using inertial sensors. J Biomech 42:2330–2335 Fioretti S, Cappozzo A, Lucchetti L (1997) Joint kinematics. In: Allard P, Cappozzo A, Lumberg A, Vaughan K (eds) Three-dimensional analysis of human locomotion. Wiley, New York, pp 173–189 Fraysse F, Thewlis D (2014) Comparison of anatomical, functional and regression methods for estimating the rotation axes of the forearm. J Biomech 47:3488–3493 Frigo C, Rabuffetti M, Kerrigan DC, Deming LC, Pedotti A (1998) Functionally oriented and clinically feasible quantitative gait analysis method. Med Biol Eng Comput 36:179–185 Gamage SSHU, Lasenby J (2002) New least squares solutions for estimating the average Centre of rotation and the axis of rotation. J Biomech 35:87–93 Garling EH, Kapteina BL, Mertens B, Barendregt W, Veeger HEJ, Nelissen RGHH, Valstar ER (2007) Soft-tissue artefact assessment during step-up using fluoroscopy and skin-mounted markers. J Biomech 40:S18–S24 Gasparutto X, Sancisi N, Jacquelin E, Parenti-Castelli V, Dumas R (2015) Validation of a multibody optimization with knee kinematic models including ligament constraints. J Biomech 48:1141–1146 Grimpampi E, Camomilla V, Cereatti A, De Leva P, Cappozzo A (2014) Metrics for describing softtissue artefact and its effect on pose, size, and shape of marker clusters. IEEE Trans Biomed Eng 61:362–367 Grood ES, Suntay WJ (1983) A joint coordinate system for the clinical description of threedimensional motions: application to the knee. J Biomech Eng 105:136–144 Gruber K, Ruder H, Denoth J, Schneider K (1998) A comparative study of impact dynamics: wobbling mass model versus rigid body models. J Biomech 31:439–444 Guan S, Gray HA, Keynejad F, Pandy MG (2016) Mobile biplane X-ray imaging system for measuring 3D dynamic joint motion during overground gait. IEEE Trans Med Imaging 35:326–336 Halvorsen K (2003) Bias compensated least square estimate of the center of rotation. J Biomech 36:999–1008 Halvorsen K, Lesser M, Lundberg A (1999) A new method for estimating the axis of rotation and the center of rotation. J Biomech 32:1221–1227 Hara R, McGinley J, Briggs C, Baker R, Sangeux M (2016) Predicting the location of the hip joint centres, impact of age group and sex. Sci Rep 6:37707 Harrington ME, Zavatsky AB, Lawson SE, Yuan Z, Theologis TN (2007) Prediction of the hip joint centre in adults, children, and patients with cerebral palsy based on magnetic resonance imaging. J Biomech 40:595–602 Kadaba MP, Ramakrishnan HK, Wootten ME (1990) Measurement of lower extremity kinematics during level walking. J Orthop Res 8:383–392 Kainz H, Carty CP, Modenese L, Boyd RN, Lloyd DG (2015) Estimation of the hip joint centre in human motion analysis: a systematic review. Clin Biomech 30:319–329
44
V. Camomilla et al.
Lamberto G, Martelli S, Cappozzo A, Mazzà C (2016) To what extent is joint and muscle mechanics predicted by musculoskeletal models sensitive to soft tissue artefacts? J Biomech. https://doi. org/10.1016/j.jbiomech.2016.07.042i Leardini A, Cappozzo A, Catani F, Toksvig-Larsen S, Petitto A, Sforza V, Cassanelli G, Giannini S (1999) Validation of a functional method for the estimation of hip joint centre location. J Biomech 32:99–103 Leardini A, Chiari L, Cappozzo A, Della Croce U (2005) Human movement analysis using stereophotogrammetry. Part 3: soft tissue artifact assessment and compensation. Gait Posture 21:212–225 Leardini A, Sawacha Z, Paolini G, Ingrosso S, Nativo R, Benedetti MG (2007) A new anatomically based protocol for gait analysis in children. Gait Posture 26:560–571 Lempereur M, Leboeuf F, Brochard S, Rousset J, Burdin V, Rémy-Néris O (2010) In vivo estimation of the glenohumeral joint center by functional methods: accuracy and repeatability assessment. J Biomech 43:370–374 Li K, Zheng L, Tashman S, Zhang X (2012) The inaccuracy of surface-measured model-derived tibiofemoral kinematics. J Biomech 45:2719–2723 Liu W, Nigg BM (2000) A mechanical model to determine the influence of masses and mass distribution on the impact force during running. J Biomech 33:219–224 Lu TW, O’Connor JJ (1999) Bone position estimation from skin marker coordinates using global optimization with joint constraints. J Biomech 32:129–134 Luinge HJ, Veltink PH, Baten CTM (2007) Ambulatory measurement of arm orientation. J Biomech 40:78–85 Masum MA, Pickering MR, Lambert AJ, Scarvell JM, Smith PN (2016) Multi-slice ultrasound image calibration of an intelligent skin-marker for soft tissue artefact compensation. J Biomech. https://doi.org/10.1016/j.jbiomech.2016.12.030 McGinnis RS, Perkins NC (2013) Inertial sensor based method for identifying spherical joint center of rotation. J Biomech 46:2546–2549 Melhem E, Assi A, El Rachkidi R, Ghanem I (2016) EOS ® biplanar X-ray imaging: concept, developments, benefits, and limitations. J Child Orthop 10:1–14 Pain MTG, Challis JH (2001) High resolution determination of body segment inertial parameters and their variation due to soft tissue motion. J Appl Biomech 17:326–334 Paul JP (1992) Terminology and units. Deliverable n. 4, C.E.C. Program AIM, Project A-2002, CAMARC-II Piazza SJ, Erdemir A, Okita N, Cavanagh PR (2004) Assessment of the functional method of hip joint center location to reduced range of hip motion. J Biomech 37:349–356 Picerno P, Cereatti A, Cappozzo A (2008) Joint kinematics estimate using wearable inertial and magnetic sensing modules. Gait Posture 28:588–595 Pierrynowski MR, Ball KA (2009) Oppugning the assumptions of spatial averaging of segment and joint orientations. J Biomech 42:375–378 Quijano S, Serrurier A, Aubert B, Laporte S, Thoreux P, Skalli W (2013) Three-dimensional reconstruction of the lower limb from biplanar calibrated radiographs. Med Eng Phys 35:1703–1712 Rabuffetti M, Ferrarin M, Mazzoleni P, Benvenuti F, Pedotti A (2003) Optimised procedure for the calibration of the force platform location. Gait Posture 17:75–80 Reinbolt JA, Schutte JF, Fregly BJ, Koh B, Haftka RT, George AD, Mitchell KH (2005) Determination of patient-specific multi-joint kinematic models through two-level optimization. J Biomech 38:621–626 Richard V, Lamberto G, Lu TW, Cappozzo A, Dumas R (2016) Knees kinematics estimation using multi-body optimisation embedding a knee joint stiffness matrix: a feasibility study. PLoS ONE. https://doi.org/10.1371/journal.pone.01570 Riddick RC, Kuo AD (2016) Soft tissues store and return mechanical energy in human running. J Biomech 49:436–441
Three-Dimensional Reconstruction of the Human Skeleton in Motion
45
Scheys L, Desloovere K, Spaepen A, Suetens P, Jonkers I (2011) Calculating gait kinematics using MR-based kinematic models. Gait Posture 33:158–164 Seel T, Schauer T, Raisch J (2012) Joint axis and position estimation from inertial measurement data by exploiting kinematic constraints. In: Proceedings of IEEE international conference on control applications, Dubrovnik, Croatia, pp 45–49 Seidel GK, Marchinda DM, Dijkers M, Soutas-Little RW (1995) Hip joint center location from palpable bony landmarks – a cadaver study. J Biomech 28:995–998 Sheehan FT (2010) The instantaneous helical axis of the subtalar and talocrural joints: a non-invasive in vivo dynamic study. J Foot Ankle Res 3:13. https://doi.org/10.1186/17571146-3-13 Sholukha V, Van Sint JS, Snoeck O, Salvia P, Moiseev F, Rooze M (2009) Prediction of joint center location by customizable multiple regressions: application to clavicle, scapula and humerus. J Biomech 42:319–324 Shuster MD (1993) A survey of attitude representations. J Astronaut Sci 41:439–517 Soderkvist I, Wedin PA (1993) Determining the movements of the skeleton using well-configured markers. J Biomech 26:1473–1477 Stagni R, Leardini A, Cappozzo A, Benedetti MG, Cappello A (2000) Effects of hip joint centre mislocation on gait analysis results. J Biomech 33:1479–1487 Van Sint JS, Hilal I, Salvia P, Sholukha V, Poulet P, Kirokoya I, Rooze M (2003) Data representation for joint kinematics simulation of the lower limb within an educational context. Med Eng Phys 25:213–220 Wakeling JM, Nigg BM (2001) Soft-tissue vibrations in the quadriceps measured with skin mounted transducers. J Biomech 34:539–543 Woltring HJ (1994) 3-D attitude representation of human joints, a standardisation proposal. J Biomech 27:1399–1414 Wu G, Cavanagh PR (1995) ISB recommendations for standardization in the reporting of kinematic data. J Biomech 28:1257–1261 Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D, Whittle M, D’Lima DD, Cristofolini L, Witte H, Schmid O, Stokes I (2002) ISB recommendation on definitions of joint coordinate system of various joints for the reporting of human joint motion – part I, ankle, hip, and spine. J Biomech 35:543–548 Yin L, Chen K, Guo L, Cheng L, Wang F, Yang L (2015) Identifying the functional flexionextension axis of the knee: an in-vivo kinematics study. PLoS ONE 10:e0128877 Zelik KE, Kuo AD (2010) Human walking isn’t all hard work: evidence of soft tissue contributions to energy dissipation and return. J Exp Biol 213:4257–4264
Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis Raphaël Dumas and Janis Wojtusch
Abstract
Body segment inertial parameters (BSIPs) of the human body are key parameters in biomechanics to study the dynamics of human motion. BSIPs can be obtained in different ways including direct measurements on cadavers or photogrammetry and medical imaging on living humans, but they are more generally estimated by regression equations (based on those measurements). This chapter overviews three widely used regression equations reported by Winter (2009), de Leva (1996a), and Dumas et al. (2007a). These regression equations are presented for the head with neck, thorax, abdomen, pelvis, and right upper arm, forearm, hand, thigh, shank, and foot segments. The segment endpoints and segment reference frames defined at the time of the BSIPs assessment and regression computation are reviewed so that the reader can consider how they match with the construction of the rigid body biomechanical models they would like to use for motion analysis. The segment definitions and regression equations that remain undefined or unavailable are indicated, and some assumptions are proposed to amend them, where found applicable. The computation of the segment mass, position of center of mass, moments, and products of inertia from these regression equations are fully detailed, including the modification of the designation of the segment axes and the transformation from right to left segments.
R. Dumas (*) LBMC UMR_T9406, Univ Lyon, Université Claude Bernard Lyon 1, IFSTTAR, Lyon, France e-mail: [email protected] J. Wojtusch Department of Computer Science, Simulation, Systems Optimization and Robotics Group, TU Darmstadt, Darmstadt, Germany e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_147
47
48
R. Dumas and J. Wojtusch
Keywords
Segment mass • Center of mass • Moments of inertia • Regression equations • Segment length • Segment endpoints • Joint center • Segment reference frame
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Computation of the BSIPs from the Regression Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Segment Definition and Regression Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48 49 50 52 74 75
Introduction Body segment inertial parameters (BSIPs) are required in biomechanics for the computation of intersegmental moments, angular momentum, mechanical work, and for the study of the whole body dynamic stability (see chapters ▶ “Induced Acceleration and Power Analyses of Human Motion”, ▶ “Optimal Control Modeling of Human Movement”, and ▶ “Physics-Based Models for Human Gait Analysis”). BSIPs refer to the segment mass, the position of the segment center of mass with respect to a segment reference frame, and the segment moments and products of inertia with respect to a segment point (typically the segment center of mass or a segment endpoint) and with respect to a segment reference frame (see chapter ▶ “Three-Dimensional Reconstruction of the Human Skeleton in Motion”). BSIPs of the human body can be obtained in different ways. The history and the description of the techniques used to assess BSIPs can be found in Pearsall and Reid 1994, Reid and Jensen 1990, and Drillis et al. 1964. BSIPs can be directly measured on cadavers (Dempster 1955; Clauser et al. 1969; Chandler et al. 1975) and indirectly measured on living subjects, typically through photogrammetry (Ackland et al. 1988; Jensen 1978; McConville et al. 1980; Young et al. 1983) or medical imaging (Bauer et al. 2007; Dumas et al. 2005; Durkin et al. 2002; Mungiole and Martin 1990; Pearsall et al. 1996; Zatsiorsky et al. 1990; Cheng et al. 2000). Based on these measurements, estimations of the BSIPs from regression equations are more classically used. Obviously, these regression equations are limited by the number and the nature of the subjects on which they have been established. In other words, regression equations for BSIPs are better established for older Caucasian non-pathological males than for females, children, and, as a matter of fact, for pathological subjects. Nevertheless, regression equations are widely used because of their expediency. Furthermore, when studying the dynamics of human motion with a 3D rigid body biomechanical model, another difficulty is to match the segment definition used for the BSIP assessment with the model construction. Indeed, BSIPs have been obtained
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . .
49
with their own rationale for segment endpoints and segment reference frames, generally constrained by experimental and technical limitations. Conversely, rigid body biomechanical models and especially “conventional gait models” widely used in clinical motion analysis (Davis et al. 1991; Kadaba et al. 1990) (see chapter ▶ “The Conventional Gait Model - Success and Limitations”), are based on a chosen marker set and calibration protocol that aims at approximating the joint centers and axes. In this context, several adjustment procedures of existing regression equations have been proposed (de Leva 1996a; Dumas et al. 2007a; Hinrichs 1990) for a better correspondence between segment definition and model construction. Still, no specific adjustment for “conventional gait models” has been proposed, and it is commonly assumed that the segment definition used for the BSIP assessment and the model construction are consistent. Therefore, classically, the segment center of mass is expected to align with the axis linking the joint centers, and this axis plus two orthogonal axes are expected to be principal axes of inertia. This chapter overviews three widely used sets of regression equations for BSIPs, allowing for a 16-segment rigid body biomechanical model. The segment definitions (i.e., segment endpoints and segment axes or planes) used for the BSIP assessment, regression computation, and adjustment are reviewed so that the reader can consider how they match with the construction of the “conventional gait models” or any rigid body biomechanical models they would like to use. The segment definitions and regression equations that remain undefined or unavailable are specified, and some assumptions are proposed to amend them, where found applicable. Note that the planes of segmentation used for the BSIP assessment can be also an issue (leading to some differences between the regression equations, especially for the trunk segments), but, conversely to the segment endpoints, axes, and planes, this does not directly interfere with the construction of the biomechanical models used in motion analysis and this is not reviewed in this chapter.
State of the Art Three widely used regression equations for BSIP are the regression equations of Winter (2009) derived from the data of Dempster (1955), the regression equations of de Leva (1996a) adjusted from the data of Zatsiorsky et al. (1990), and the regression equations of Dumas et al. (2007a, b) adjusted from the data of McConville et al. (1980) and Young et al. (1983). Dempster (1955) directly measured the BSIPs of eight male cadavers (mean age 68.5 years old, mean weight 61.1 kg, mean stature 1.69 m) using equilibrium and pendulum methods. Zatsiorsky et al. (1990) indirectly measured the BSIPs by frontal gamma-ray scanner on 100 males (mean age 23.8 years old, mean weight 73.0 kg, mean stature 1.74 m) and 15 females (mean age 19.0 years old, mean weight 61.9 kg, mean stature 1.73 m). They obtained the surface density of the body from subjects lying supine. The foot segment was scanned in a lateral view
50
R. Dumas and J. Wojtusch
separately from the rest of the body. The whole human body was modeled as rectangular cuboids of 2 cm width and length and of different heights estimated from anthropometric measurements. The masses and the distances from the geometrical centers of the cuboids to reference anatomical landmarks were known from the surface density. The centers of the mass of the cuboids were assumed at their geometrical centers and the principal axes of inertia of the cuboids were assumed aligned with the axes of symmetry of the cuboids. Then, the BSIPs of the segments were computed by summing up the BSIPs of the cuboids (i.e., using weighted barycenter and parallel axis theorem). McConville et al. (1980) and Young et al. (1983) indirectly measured the BSIPs by photogrammetry on 31 males (mean age 27.5 years old, mean weight 77.3 kg, mean stature 1.77 m) and 46 females (mean age 31.2 years old, mean weight 63.9 kg, mean stature 1.61 m), respectively. The BSIPs are defined relative to skin anatomical landmarks assuming a homogenous density of 1 g/cm3. Based on these datasets and according to the aforementioned adjustments, linear regression equations have been proposed (de Leva 1996a; Dumas et al. 2007a; Winter 2009). The segment mass is computed as a percentage of the body mass. The position of the center of mass is computed as a percentage of the segment length, defined as the distance between the segment endpoints. The radii of gyration (i.e., the square roots of the moments of inertia divided by the segment mass) are computed as percentages of the segment length. In these regression equations, the number of segments is 16: the head with neck, thorax, abdomen, pelvis, right and left upper arms, forearms and hands, and thighs, shanks, and feet. In the literature, other segmentations and other regression equations (i.e., nonlinear (Zatsiorsky et al. 1990), involving anthropometric measurements such as segment circumferences (Yeadon and Morlock 1989; Zatsiorsky et al. 1990)) exist but appear hardly used, probably because they do not fit well with the motion analysis protocols (i.e., they involve not only skin markers but also calipers and tape measures).
Computation of the BSIPs from the Regression Equations As previously mentioned, the segment mass, ms, of segment s = 1,. . .,16 is estimated as: ms ¼ ps M
(1)
where ps is the percentage of the body mass M. The position of the center of mass with respect to the segment reference frame is estimated as: 0
1 cXs A rs ¼ Ls @ cY s Z cs
(2)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . .
51
Z where cXs , cY s , cs are coordinates of the center of mass expressed as percentages of the segment length Ls. The inertia matrix (moments and products of inertia in and out the diagonal, respectively) with respect to the center of mass and the segment reference frame is estimated as:
2 6 Is ¼ ms ðLs Þ2 4
r XX s
2
XY 2 r sYY 2 rs
sym:
XZ 2 3 r sYZ 2 7 5 r sZZ 2 rs
(3)
YY ZZ where r XX s , r s , r s are radii of gyration (i.e., the square roots of the moments of inertia divided by the segment mass) expressed as percentages of the segment length Ls. The products of inertia are expressed in the same way (i is indicated in case of negative products of inertia). The designation of the segment axes can be different from one rigid body biomechanical model to another. The regression equations in the following tables will be labeled with respect to the anterior-posterior, superior-inferior, and mediallateral axes. These labels are intended for a subject in anatomic posture (standing upright with arms at the sides, palms facing forward, and feet parallel). Note that in the “conventional gait models,” I axis is anterior, J axis is lateral (to the left), and K axis is superior (Davis III et al. 1991; Kadaba et al. 1990). However, according to the standardization of the International Society of Biomechanics (ISB), X axis is anterior, Y axis is superior, and Z axis is lateral (to the right) (Wu et al. 2002, 2005). The sign convention used in the following tables complies with this standardization of the ISB (X, Y, and Z axes). Nevertheless, the position of the center of mass and the inertia matrix with respect to the I, J, and K axes can be easily computed using a permutation matrix P:
# 0 X1 cs 1 0 0 A rs ¼ 0 0 1 L s @ cY s 0 1 0 Z cs |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} "
(4)
P
and 2 6 Is ¼ ms ðLs Þ2 ½P4
r XX s
2
sym:
XY 2 r sYY 2 rs
XZ 2 3 r sYZ 2 7 T 5 ½ P : r sZZ 2 rs
(5)
In the same way, the regression equations in the following tables are for the right upper and lower limb segments. The position of the center of mass and the inertia matrix for the left segments can be computed (in the X, Y, and Z axes) using a symmetry matrix S (in place of P) in Eqs. 4 and 5:
52
R. Dumas and J. Wojtusch
2
1 S ¼ 40 0
3 0 0 1 0 5: 0 1
(6)
The inertia matrix (moments and products of inertia) in Eq. 3 is given with respect to the center of mass and the segment reference frame. The inertia matrix with respect to the origin of the segment reference frame can be computed with the parallel axis theorem: Is ¼ Is þ m s
ðrs ÞT rs E33 rs ðrs ÞT
(7)
where E3 3 is the identity matrix.
Segment Definition and Regression Equations For the regression equations of Winter (2009), the BSIPs for the segments of the trunk are revised based on the original data of Dempster (1955) and the additional estimations of Plagenhoef (1971): the center of mass of the abdomen and pelvis segments was estimated by equilibrium methods applied to a cardboard-lead cutout modeling the anthropometry and mass distribution of the trunk as reported by Dempster (1955). Note that, in the original data of Dempster (1955), only an abdominopelvic segment is reported. Its mass is 26.4% of body mass and its segment density is 1.01 g/cm3. The position of its center of mass about the superior-inferior axis is 59.9% of segment length (i.e., thoracic joint center to midpoint between hip joint centers). The abdomen and pelvis segments are presented in the following tables for consistency among the different regression equations. Nevertheless, this abdominopelvic segment may be preferred if this segmentation matches with the rigid body biomechanical model used for motion analysis. Moreover, the abdomen and pelvis segments were also available in the regression equations reported by Winter (2009), but the position of the center of mass of these segments, as well as the thorax segment, was given in percentage of the longitudinal distance between midpoints between glenohumeral and hip joint centers (i.e., 30.4% of total height (Plagenhoef 1971)). As for the regression equations based on the data of McConville et al. (1980) and Young et al. (1983), the regression equations provided in the following tables include updates for the thorax and abdomen segments (Dumas et al. 2015) and a revision made for this chapter (i.e., some typos and inconsistency in the regression equations for the joint centers previously used in the adjustment procedure were corrected). The segment endpoints and segment axes are presented in Fig. 1 with the related skin landmarks and joint centers estimated by different regression equations. The regression equations for the BSIPs of the different segments are given in (Tables 1–10).
3.2%
0.6%
Gleno-humeral joint centre
Hip joint centre
Tip of 2nd toe
1rst metatarsal head
Heel
Medial maleollus Shyrion
100%
Knee joint centre Tibiale
7.4% Medial femoral epicondyle
0.7%
3rd metacarpale
5th metacarpal heads
Ulnar styloid
Wrist joint centre
100%
4.3% Medial humeral epicondyle Elbow joint centre
10.4%
12th thoracic vertebra
8th thoracic vertebra
Acromion
7th cervical vertebra
Head vertex
Hip joint centre
Right anterior-superior iliac spine
3.2% 37.0% 36.1%
Midpoint between posterior-superior iliac spines Lumbar joint centre
Fig. 1 Skin landmarks, joint centers estimated by regression equations, and body segments
5th metatarsal head
Ankle joint centre
Lateral maleollus
Fibula head
Lateral femoral epicondyle
100%
Greater trochanter
2nd metacarpal heads
Radius styloid
Lateral humeral epicondyle Radiale
100%
Acromion
8°
94°
100%
33.5%
52%
33% 11°
55%
9.5%
100%
Left anterior-superior iliac spine
Thoracic joint centre
Xiphoid
Gleno-humeral joint centre
Suprasternal notch
Cervical joint centre
Sellion
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 53
(Dempster 1955; Winter 2009) Head vertex skin landmarka Cervical joint center obtained by dissection (center of C7-T1 disc)
Superior-inferior axis from cervical joint center to head vertex skin landmark Sagittal plane set as plane of oscillation of the pendulum method Origin at head vertex skin landmarka
The head with neck Segment endpoints
Segment reference frame
Table 1 Head with neck
Superior-inferior axis: N/A Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at head vertex skin landmark projected on superior-inferior axis
(de Leva 1996a; Zatsiorsky et al. 1990) Skin landmark on head vertex projected on superiorinferior axis Skin landmark on 7th cervical vertebra projected on superior-inferior axis
(Dumas et al. 2007b; McConville et al. 1980; Young et al. 1983) Skin landmark on the head vertex Cervical joint center estimated by the regression equations (Dumas et al. 2007a) Female: cervical joint center on a direction forming an angle of 14 in the sagittal plane with the vector from 7th cervical vertebra to suprasternal notch skin landmarks and at 53% of the thorax width (i.e., distance between 7th cervical vertebra and suprasternal notch skin landmarks) from 7th cervical vertebra skin landmark Male: angle of 8 and percentage of 55% Superior-inferior axis from cervical joint center to head vertex skin landmark Sagittal plane containing cervical joint center and skin landmarks on the head vertex and sellion Origin at cervical joint center
54 R. Dumas and J. Wojtusch
a
Female: 0.1 and male: 0.1 Female: 30 and male: 28 Female: 24 and male: 21
Female: 26.1 and male: 26.1 Female: 27.1 and male: 30.3 Assumed negligible Assumed negligible Assumed negligible
Assumed negligible Assumed equal to mediallateral axis N/A Female: N/A and male: 28.1a Assumed negligible Assumed negligible Assumed negligible
Value of Winter (2009) replaced by original value of Dempster (1955), radius of gyration worked out with a rule of 3
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Female: 0 and male: 3
Female: 31 and male: 30 Female: 5(i) and male: 7(i) Female: 1 and male: 2(i)
Female: 55.9 and male: 53.4
Female: 48.41 and male: 50.02 Assumed negligible Female: 29.5 and male: 31.5
Female: N/A and male: 43.3a
Female: 0.8 and male: 2.0
Assumed to be 1 Female: 6.7 and male: 6.7
Female: 243 and male: 278
N/A
Female: 243.7 and male: 242.9 N/A Female: 6.68 and male: 6.94
Assumed negligible
1.11 Female: N/A and male: 8.10
Segment density (g/cm3) Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Anterior-posterior axis (X) Superior-inferior axis (Y) Medial-lateral axis (Z) Anterior-posterior axis (X) Superior-inferior axis (Y) Medial-lateral axis (Z) Sagittal plane (X,Y) Transverse plane (X, Z) Frontal plane (Y,Z)
N/A
Segment length (mm)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 55
Segment reference frame
The thorax Segment endpoints
Table 2 Thorax
Superior-inferior axis from thoracic to cervical joint center Sagittal plane set as plane of oscillation of the pendulum method Origin at cervical joint center
(Dempster 1955; Winter 2009) Cervical joint center (see head with neck segment) Thoracic joint center obtained by dissection (center of T12-L1 disc)
Superior-inferior axis: N/A Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at 7th cervical vertebra skin landmark projected on superior-inferior axis
(de Leva 1996a; Zatsiorsky et al. 1990) Skin landmark on the seventh cervical vertebra projected on superior-inferior axis Skin landmark on xiphoid projected on superior-inferior axis
(Dumas et al. 2015; McConville et al. 1980; Young et al. 1983) Cervical joint center (see head with neck segment) Thoracic joint center estimated by the regression equations (Dumas et al. 2015) Female: thoracic joint center on a direction forming an angle of 92 in the sagittal plane with the vector from 12th to 8th thoracic vertebra skin landmarks and at 50% of thorax width (see head with neck segment) from 12th thoracic vertebra skin landmark Male : angle of 94 and percentage of 52% Superior-inferior axis from thoracic to cervical joint center Sagittal plane containing skin landmarks on 7th cervical and 8th thoracic vertebra and suprasternal notch Origin at cervical joint center
56 R. Dumas and J. Wojtusch
a
Anterior-posterior axis (X) Superior-inferior axis (Y) Medial-lateral axis (Z) Anterior-posterior axis (X) Superior-inferior axis (Y) Medial-lateral axis (Z) Sagittal plane (X, Y) Transverse plane (X,Z) Frontal plane (Y, Z) Assumed negligible Assumed negligible
Assumed negligible Assumed negligible
Female: 46.6 and male: 50.5
N/A N/A
Female: 44.9 and male: 46.5
N/A
N/A
Female: 31.4 and male: 32.0
N/A
Female: 54.2 and male: 55.5
Female: 50.50 and male: 50.66 Assumed negligible
Female: N/A and male: 62.7a Assumed negligible
Female: 1 and male: 3
Female: 3(i) and male: 1
Female: 12(i) and male: 11(i)
Female: 34 and male: 36
Female: 32 and male: 33
Female: 38 and male: 42
Female: 0.1 and male: 0.4
Female: 1.5 and male: 0.0
Female: 322 and male: 334 Assumed to be 1 Female: 26.3 and male: 30.4
N/A
Female: 228.0 and male: 242.1 N/A Female: 15.45 and male: 15.96
Assumed negligible
N/A 0.92 Female: N/A and male: 21.60
Value of Winter (2009) replaced by original value of Dempster (1955)
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 57
Segment reference frame
The abdomen Segment endpoints
Table 3 Abdomen
Superior-inferior axis: N/A Sagittal plane of the cardboard-lead cutouts modeling the anthropometry and mass distribution of the trunk Origin at thoracic joint centera
(Dempster 1955; Plagenhoef 1971; Winter 2009) Thoracic joint center (see thorax segment) Inferior endpoint: N/A
Superior-inferior axis: N/A Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at xiphoid skin landmark projected on superior-inferior axis
(de Leva 1996a; Zatsiorsky et al. 1990) Skin landmark on xiphoid projected on superior-inferior axis Skin landmark on omphalion projected on superior-inferior axis
(Dumas et al. 2015; McConville et al. 1980; Young et al. 1983) Thoracic joint center (see thorax segment) Lumbar joint center estimated by the regression equations (Dumas et al. 2007a) Female: in the pelvis reference frame (see pelvis segment) with origin translated at midpoint between anterior-superior iliac spine skin landmarks, lumbar joint center at 34.0%, 4.9% and 0% of pelvis width (see pelvis segment) about the anteriorposterior, superior-inferior, and mediallateral axes, respectively Male: percentages of 33.5%%, 3.2%, and 0.0%, respectively Superior-inferior axis from thoracic to lumbar joint center Sagittal plane: N/A (no axial rotation at lumbar joint center assumed) Origin at thoracic joint center
58 R. Dumas and J. Wojtusch
N/A N/A
Superior-inferior axis (Y)
Medial-lateral axis (Z) N/A Assumed negligible Assumed negligible
Assumed negligible N/A
Medial-lateral axis (Z) Anterior-posterior axis (X)
Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z)
N/A Female: N/A and male: 34.6a
Anterior-posterior axis (X) Superior-inferior axis (Y)
N/A Female: 45.12 and male: 45.02 Assumed negligible Female: 35.4 and male: 38.3 Female: 41.5 and male: 46.8 Female: 43.3 and male: 48.2 N/A Assumed negligible Assumed negligible
Female: 14.65 and male: 16.33
Female: N/A and male: 13.90
N/A
Female: 205.3 and male: 215.5 N/A
N/A
Female: 25 and male: 11 Female: 3(i) and male: 6(i) Female: 5(i) and male: 5(i)
Female: 52 and male: 40
Female: 78 and male: 66
Female: 0.3 and male: 3.3 Female: 65 and male: 54
Female: 21.9 and male: 17.6 Female: 41.0 and male: 36.1
Female: 4.1 and male: 2.9
Assumed to be 1
Female: 125 and male: 151
a Adapted from Plagenhoef (1971), position of center of mass expressed as percentage of the length of the abdominopelvic segment (thoracic joint center to the midpoint between the hip joint centers)
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 59
Female: 181.5 and male: 145.7 N/A
N/A
N/A
Segment length (mm) Segment density (g/cm3)
Superior-inferior axis: N/A Frontal plane parallel to the gammaray scanner acquisition plane Origin at omphalion skin landmark projected on superior-inferior axis
(de Leva 1996a; Zatsiorsky et al. 1990) Skin landmark on omphalion projected on superior-inferior axis Midpoint between hip joint centers estimated by the regression equations (de Leva 1996b) Female: N/A Male: on superior-inferior axis, hip joint center at 0.7% of thigh length (i.e., longitudinal distance between tibial and greater trochanter skin landmarks) from the greater trochanter skin landmark (percentages on anterior-posterior and media-lateral axes: N/A)
Superior-inferior axis: N/A Sagittal plane of the cardboard-lead cutouts modeling the anthropometry and mass distribution of the trunk Origin at midpoint between hip joint centers
(Dempster 1955; Plagenhoef 1971; Winter 2009) Superior endpoint: N/A Midpoint between hip joint centers obtained by dissection (acetabulum center)
Segment reference frame
The pelvis Segment endpoints
Table 4 Pelvis
Assumed to be 1
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Lumbar joint center (see abdomen segment) Midpoint between hip joint centers estimated by the regression equations (Dumas et al. 2007a) Female: in the pelvis reference frame (see below) with origin translated at midpoint between anterior-superior iliac spine skin landmarks, right/left hip joint center at 13.9%, 33.6%, and +/37.2% of pelvis width (i.e., distance between left and right anterior-superior iliac spine skin landmarks) about the anteriorposterior, superior-inferior, and medial-lateral axes, respectively Male: percentages of 9.5%, 37.0%, and +/36.1%, respectively Medial-lateral axis from left to right anterior superior iliac spine skin landmarks Transverse plane containing skin landmarks on left and right anteriorsuperior iliac spines and midpoint between posterior-superior iliac spines Origin at lumbar joint center Female: 103 and male: 93
60 R. Dumas and J. Wojtusch
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z) Assumed negligible Assumed negligible
Assumed negligible
Assumed negligible
Female: 43.3 and male: 61.5
N/A
N/A
Female: 44.4 and male: 58.7
N/A
N/A
Female: 40.2 and male: 55.1
Assumed negligible
Assumed negligible
N/A
Female: 22.8 and male: 28.2
Female: 49.20 and male: 61.15
Female: N/A and male: 15.6a
Female: 2(i) and male: 8(i)
Female: 3(i) and male: 12(i)
Female: 35(i) and male: 25(i)
Female: 82 and male: 96
Female: 105 and male: 106
Female: 95 and male: 102
Female: 0.2 and male: 0.6
Female: 7.2 and male: 0.2
Female: 14.7 and male: 14.2
N/A
Female: 12.47 and male: 11.17
N/A
Female: N/A and male: 14.20
a Adapted from Plagenhoef (1971), position of center of mass expressed as percentage of the length of the abdominopelvic segment (thoracic joint center to the midpoint between the hip joint centers)
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 61
Superior-inferior axis from elbow to glenohumeral joint center Frontal plane parallel to the gammaray scanner acquisition plane Origin at glenohumeral joint center
Female: 275.1 and male: 281.7 N/A
Female: N/A and male: 286a
1.07
Segment length (mm) Segment density (g/cm3)
(de Leva 1996a; Zatsiorsky et al. 1990) Glenohumeral and elbow joint centers estimated by the regression equations (de Leva 1996b) Female: N/A Male: on superior-inferior axis, glenohumeral joint center at 10.4% of upper arm length (i.e., longitudinal distance between acromion and radial skin landmarks) from acromion skin landmark and elbow joint center at 4.3% of upper arm length from radial skin landmark (percentages on anterior-posterior and media-lateral axes: N/A)
Superior-inferior axis from elbow to glenohumeral joint center Sagittal plane set as plane of oscillation of the pendulum method Origin at glenohumeral joint center
(Dempster 1955; Plagenhoef 1971; Winter 2009) Glenohumeral joint center obtained by dissection (center of curvature of humeral head) Elbow joint center obtained by dissection (axis of humeral trochlea at narrowest cross section of ulnar articulation)
Segment reference frame
The upper arm Segment endpoints
Table 5 Upper arm
Assumed to be 1
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Glenohumeral joint center estimated by the regression equations (Dumas et al. 2007a) Elbow joint center estimated at midpoint between lateral and medial humeral epicondyle skin landmarks Female : glenohumeral joint center on a direction forming an angle of 5 in the sagittal plane with the vector from 7th cervical vertebra to suprasternal notch skin landmarks and at 36% of thorax width (see head with neck and thorax segments) from the acromion skin landmark Male: angle of 11 and percentage of 33% Superior-inferior axis from elbow to glenohumeral joint center Frontal plane containing glenohumeral joint center and skin landmarks on lateral and medial humeral epicondyles Origin at glenohumeral joint center Female: 251 and male: 277
62 R. Dumas and J. Wojtusch
a
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X, Y) Transverse plane (X,Z) Frontal plane (Y,Z) Assumed negligible
Assumed negligible Assumed negligible
Assumed negligible
Assumed negligible
Female: 27.8 and male: 28.5
Female: N/A and male: 32.2
Assumed negligible
Female: 14.8 and male: 15.8
N/A
Assumed negligible
Assumed negligible
Female: 26.0 and male: 26.9
Female: 57.54 and male: 57.72
Female: N/A and male: 43.6
Assumed equal to medial-lateral axis
Assumed negligible
Female: 2.55 and male: 2.71
Assumed negligible
Female: N/A and male: 2.80
Value of Plagenhoef (1971), 16.9% of total height
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Female: 3 and male: 13(i)
Female: 5 and male: 3
Female: 3(i) and male: 5
Female: 30 and male: 30
Female: 15 and male: 13
Female: 30 and male: 29
Female: 3.3 and male: 3.1
Female: 50.0 and male: 48.2
Female: 5.5 and male: 1.8
Female: 2.3 and male: 2.4
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 63
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass)
Segment reference frame
The forearm Segment endpoints
Table 6 Forearm
Female: N/A and male: 1.60
(Dempster 1955; Plagenhoef 1971; Winter 2009) Elbow joint center (see upper arm segment) Wrist joint center obtained by dissection (center of curvature of proximal end of capitate bone) Superior-inferior axis from wrist to elbow joint center Sagittal plane set as plane of oscillation of the pendulum method Origin at elbow joint center Female: N/A and male: 269a 1.13 Assumed to be 1
N/A
Female: 1.4 and male: 1.7
Female: 247 and male: 283
Female: 264.3 and male: 268.9
Female: 1.38 and male: 1.62
Superior-inferior axis from wrist to elbow joint center Frontal plane containing elbow joint center and skin landmarks on radial and ulna styloids Origin at elbow joint center
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Elbow joint center (see upper arm segment) Wrist joint center estimated at midpoint between radial and ulna styloid skin landmarks
(de Leva 1996a; Zatsiorsky et al. 1990) Elbow joint center (see upper arm segment) Wrist joint center estimated by the regression equations (de Leva 1996b) Female: N/A Male: on superior-inferior axis, wrist joint center at 0.6% of forearm length (i.e., longitudinal distance between radial and radius styloid skin landmarks) from radius styloid skin landmark (percentages on anterior-posterior and media-lateral axes: N/A) Superior-inferior axis from wrist to elbow joint center Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at elbow joint center
64 R. Dumas and J. Wojtusch
a
Female: 10 and male: 8 Female: 3 and male: 1(i) Female: 13(i) and male: 2
Female: N/A and Female: 26.1 and male: 27.6 male: 30.3 Assumed negligible Assumed negligible
Assumed negligible Assumed negligible
Assumed negligible Assumed negligible
Female: 25 and male: 28
Female: 14 and male: 11
N/A
Female: 9.4 and male: 12.1
Female: 27 and male: 28
Assumed equal to Female: 25.7 and male: 26.5 medial-lateral axis
Female: 41.1 and male: 41.7
Female: 1.9 and male: 1.1
Female: 45.59 and male: 45.74
Female: 2.1 and male: 1.3
Assumed negligible Assumed negligible
Female: N/A and male: 43.0
Assumed negligible Assumed negligible
Value of Plagenhoef (1971), 15.9% of total height
Anteriorposterior axis (X) Superiorinferior axis (Y) Medial-lateral axis (Z) Moment of Anteriorinertia (radius posterior axis of gyration in % (X) of the segment Superiorlength) inferior axis (Y) Medial-lateral axis (Z) Product of Sagittal plane inertia (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z)
Position of center of mass (% of the segment length)
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 65
Female: 76.8 and male: 83.9
Female: 74.74 and male: 79.00 Assumed negligible
Female: N/A and male: 50.6 Assumed negligible
Female: 4.8 and male: 7.5
Female: 7.7 and male: 8.2
Superior-inferior axis from midpoint between 2nd and 5th metacarpal head skin landmarks to wrist joint center Frontal plane containing wrist joint center and skin landmarks on 2nd and 5th metacarpal heads Origin at wrist joint center Female: 71 and male: 80 Assumed to be 1 Female: 0.5 and male: 0.6
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Wrist joint center (see forearm segment) Midpoint between 2nd and 5th metacarpal head skin landmarks
N/A
Female: 78.0 and male: 86.2 N/A Female: 0.56 and male: 0.61
(de Leva 1996a; Zatsiorsky et al. 1990) Wrist joint center (see forearm segment) 3rd metacarpal skin landmark projected on superior-inferior axis Superior-inferior axis: N/A Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at wrist joint center
N/A
N/A 1.17 Female: N/A and male: 0.60
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z)
Superior-inferior axis: N/A Sagittal plane set as plane of oscillation of the pendulum method Origin at wrist joint center
(Dempster 1955; Winter 2009) Wrist joint center (see forearm segment) Interphalangeal knuckle of 3rd finger skin landmark
Segment reference frame
The hand Segment endpoints
Table 7 Hand
66 R. Dumas and J. Wojtusch
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z) Assumed negligible Assumed negligible
N/A N/A
Female: 53.1 and male: 62.8
Female: N/A and male: 29.7
Assumed negligible
Female: 33.5 and male: 40.1
N/A
N/A
Female: 45.4 and male: 51.3
N/A
Female: 28(i) and male: 20(i)
Female: 23 and male: 15
Female: 29 and male: 22
Female: 59 and male: 56
Female: 43 and male: 38
Female: 64 and male: 61
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 67
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass)
Segment reference frame
The thigh Segment endpoints
Table 8 Thigh
Superior-inferior axis from knee to hip joint center Sagittal plane set as plane of oscillation of the pendulum method Origin at hip joint center Female: N/A and male: 395a 1.05 Female: N/A and male: 10.00
(Dempster 1955; Plagenhoef 1971; Winter 2009) Hip joint center obtained by dissection (center of curvature of femoral head) Knee joint center obtained by dissection (middle of a line through the center of curvature of the posterior aspect of femoral condyles)
(de Leva 1996a; Zatsiorsky et al. 1990) Hip joint center (see pelvis segment) Knee joint center estimated by the regression equations (de Leva 1996b) Female: N/A Male: on superior-inferior axis, knee joint center at 7.4% of thigh length (i.e., longitudinal distance between greater trochanter and tibial skin landmarks) from tibial skin landmark (percentages on anterior-posterior and medialateral axes: N/A) Superior-inferior axis from knee to hip joint center Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at hip joint center Female: 368.5 and male: 422.2 N/A Female: 14.78 and male: 14.16
Superior-inferior axis from knee to hip joint center Frontal plane containing hip joint center and skin landmarks on lateral and medial femoral epicondyles Female: 379 and male: 432 Assumed to be 1 Female: 14.6 and male: 12.3
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Hip joint center (see pelvis segment) Knee joint center estimated at midpoint between midpoint between lateral and medial femoral epicondyle skin landmarks
68 R. Dumas and J. Wojtusch
a
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z) Assumed negligible Assumed negligible Assumed negligible
Assumed negligible Assumed negligible
Female: 36.9 and male: 32.9
Female: N/A and male: 32.3
Assumed negligible
Female: 16.2 and male: 14.9
N/A
Assumed negligible
Assumed negligible
Female: 36.4 and male: 32.9
Female: 36.12 and male: 40.95
Female: N/A and male: 43.3
Assumed equal to medial-lateral axis
Assumed negligible
Assumed negligible
Value of Plagenhoef (1971), 23.4% of total height
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Position of center of mass (% of the segment length)
Female: 7(i) and male: 7(i)
Female: 2 and male: 2(i)
Female: 7(i) and male: 7
Female: 32 and male: 30
Female: 19 and male: 15
Female: 31 and male: 29
Female: 0.8 and male: 3.3
Female: 37.7 and male: 42.9
Female: 7.7 and male: 4.1
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 69
Female: 438.6 and male: 440.3 N/A Female: 4.81 and male: 4.33
Superior-inferior axis from ankle to knee joint center Sagittal plane set as plane of oscillation of the pendulum method Origin at knee joint center
Female: N/A and male: 428a
1.09
Female: N/A and male: 4.65
Segment length (mm) Segment density (g/cm3) Segment mass (% of total body mass)
(Dempster 1955; Plagenhoef 1971; Winter 2009) Knee joint center (see thigh segment) Ankle joint center obtained by dissection (center of the area of the cut body of the talus)
Segment reference frame
The shank Segment endpoints
(de Leva 1996a; Zatsiorsky et al. 1990) Knee joint center (see thigh segment) Ankle joint center estimated by the regression equations (de Leva 1996b) Male: on superior-inferior axis, ankle joint center at 3.2% of shank length (i.e., longitudinal distance between tibial and sphyrion skin landmarks) from sphyrion skin landmark (percentages on anterior-posterior and media-lateral axes: N/A) Superior-inferior axis from ankle to knee joint center Frontal plane parallel to the gammaray scanner acquisition plane Origin at knee joint center
Table 9 Shank
Female: 4.5 and male: 4.8
Assumed to be 1
Superior-inferior axis from ankle to knee joint center Frontal plane containing knee and ankle joint centers and the fibula head skin landmark Origin at knee joint center Female: 388 and male: 433
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Knee joint center (see thigh segment) Ankle joint center estimated at midpoint between lateral and medial malleolus skin landmarks
70 R. Dumas and J. Wojtusch
a
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z) Assumed negligible Assumed negligible Assumed negligible
Assumed negligible
Assumed negligible
Female: 26.7 and male: 25.1
Female: N/A and male: 30.2
Assumed negligible
Female: 9.2 and male: 10.2
N/A
Assumed negligible
Assumed negligible
Female: 26.3 and male: 24.6
Female: 43.52 and male: 43.95
Female: N/A and male: 43.3
Assumed equal to medial-lateral axis
Assumed negligible
Assumed negligible
Value of Plagenhoef (1971), 25.3% of total height
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Position of center of mass (% of the segment length)
Female: 6 and male: 4
Female: 1 and male: 2(i)
Female: 2 and male: 4(i)
Female: 28 and male: 28
Female: 10 and male: 10
Female: 28 and male: 28
Female: 3.1 and male: 0.7
Female: 40.4 and male: 41.0
Female: 4.9 and male: 4.8
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 71
Segment density (g/cm3) Segment mass (% of total body mass) Position of center of mass (% of the segment length)
Segment length (mm)
Segment reference frame
The foot Segment endpoints
Table 10 Foot
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z)
Female: 40.14 and Male: 44.15 Assumed negligible
Assumed negligible
Assumed negligible
Assumed negligible
Female: 228.3 and male: 258.1 N/A Female: 1.29 and male: 1.37
(de Leva 1996a; Zatsiorsky et al. 1990) Heel skin landmark Tip of longest toe skin landmark Anterior-posterior axis from heel to tip of longest toe skin landmarks Frontal plane parallel to the gamma-ray scanner acquisition plane Origin at heel skin landmark
Female: N/A and male: 42.9a
1.09 Female: N/A and male: 1.45
(Dempster 1955; Winter 2009) Heel skin landmarksa Tip of 2nd toe skin landmarksa Anterior-posterior axis from heel to tip of 2nd toe skin landmarks Sagittal plane set as plane of oscillation of the pendulum method Origin at heel skin landmarka N/A
Female: 5.5 and male: 3.4
Female: 30.9 and male: 19.9
Female: 38.2 and male: 50.2
Assumed to be 1 Female: 1.0 and male: 1.2
Female: 117 and male: 139
(Dumas et al. 2007a; McConville et al. 1980; Young et al. 1983) Ankle joint center (see shank segment) Midpoint between 1st and 5th metatarsal head skin landmarks Anterior-posterior axis from heel to midpoint between 1st and 5th metatarsal head skin landmarks Transverse plane containing skin landmarks on heel and 1st and 5th metatarsal heads
72 R. Dumas and J. Wojtusch
a
Anteriorposterior axis (X) Superiorinferior axis (Y) Mediallateral axis (Z) Sagittal plane (X,Y) Transverse plane (X,Z) Frontal plane (Y,Z) N/A Assumed negligible
N/A Assumed negligible
Female: 29.9 and male: 25.7
Female: N/A and male: 40.7a
N/A
Female: 27.9 and male: 24.5
Assumed equal to mediallateral axis
N/A
Female: 13.9 and male: 12.4
N/A
Value of Winter (2009) replaced by original value of Dempster (1955), radius of gyration worked out with a rule of 3
Product of inertia
Moment of inertia (radius of gyration in % of the segment length)
Female: 5(i) and male: 0
Female: 9 and male: 11(i)
Female: 15(i) and male: 17
Female: 50 and male: 48
Female: 50 and male: 49
Female: 24 and male: 22
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . . 73
74
R. Dumas and J. Wojtusch
Future Directions As previously mentioned in the introduction, regression equations (specifically regression equations based on body mass and segment lengths) are widely used because of their expediency but remain limited and sometimes difficult to harmonize with the rigid body biomechanical models used for motion analysis. The adaptation of the previously developed adjustment procedures to elderly subjects has been recently proposed (Ho Hoang and Mombaur 2015) and may be extended to other specific populations. There is an increasing interest in using subject-specific BSIPs of the lower limb indirectly measured by medical imaging (Bauer et al. 2007; Dao et al. 2012; Ganley and Powers 2004; Sreenivasa et al. 2016; Taddei et al. 2012; Valente et al. 2014) for gait analysis, especially in case of pathologic subjects (see chapter ▶ “Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion”). Here again, the issue of matching segment definition used for the BSIP assessment with the biomechanical model used for motion analysis exists. The segment definition and model construction are generally assumed consistent or anatomical landmarks are virtually palpated on the MRI or CT scan reconstructions (Dao et al. 2012; Sreenivasa et al. 2016; Valente et al. 2014). Occasionally, skin markers are placed on the subject before CT scan and gait analysis, both performed consecutively (Taddei et al. 2012). Nevertheless, the main purpose of these cumbersome personalization methods was generally focussed on bone and muscle geometry. The introduction of skin markers used for motion analysis within the procedure of the BSIPs assessment was more widely adopted with photogrammetry (Davidson et al. 2008; Pillet et al. 2010; Verriest 2012). This means that calibrated photographs are taken during a static posture (just before or after motion analysis) allowing to build a 3D volume model of the subject, straightforwardly registered with the rigid body biomechanical model. Moreover, without additional calibrated photographs and not performed during a static posture but a dynamic movement, the BSIPs can be estimated by identification methods (Ayusawa et al. 2014; Jovic et al. 2016; Vaughan et al. 1982). In this case, a rigid body biomechanical model and the equations of motion are directly used to compute the BSIPs that minimize the errors between model-derived and measured ground reaction forces and moments. However, as for the functional calibration of the joint centers and axes, these methods may require dedicated movements. With that respect, a method that simultaneously identifies BSIPs, joint centers, and segment lengths has been recently proposed (Bonnet et al. 2017). This method, based on extended Kalman filters, minimizes the errors between model-derived and measured ground reaction forces and moments as well as skin marker trajectories. Other advanced methods for inverse dynamics (i.e., residual elimination/reduction algorithms) typically alter the generalized accelerations of the model but may also include either some (Delp et al. 2007) or all of the BSIPs (Jackson et al. 2015) as design variables of the minimization process.
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . .
75
Most of the abovementioned methods for BSIPs assessment involve numerous skin markers, especially when all the segments of the human body (e.g., the 16 segments previously mentioned) are of interest. Therefore, another direction is to reduce the number of skin markers at its minimum while estimating the center of mass of the human body, typically for the analysis of the dynamic stability. Adapted segment definition, regression equations, and marker set have been dedicated for such specific applications (Tisserand et al. 2016; Yang and Pai 2014).
References Ackland TR, Blanksby BA, Bloomfield J (1988) Inertial characteristics of adolescent male body segments. J Biomech 21(4):319–327. https://doi.org/10.1016/0021-9290(88)90261-8 Ayusawa K, Venture G, Nakamura Y (2014) Identifiability and identification of inertial parameters using the underactuated base-link dynamics for legged multibody systems. Int J Robot Res 33(3):446–468. https://doi.org/10.1177/0278364913495932 Bauer JJ, Pavol MJ, Snow CM, Hayes WC (2007) MRI-derived body segment parameters of children differ from age-based estimates derived using photogrammetry. J Biomech 40(13):2904–2910. https://doi.org/10.1016/j.jbiomech.2007.03.006 Bonnet V, Dumas R, Cappozzo A, Joukov V, Daune G, Kulić D, Fraisse P, Andary S, Venture G (2017) A constrained extended Kalman filter for the optimal estimate of kinematics and kinetics of a sagittal symmetric exercise. J Biomech. https://doi.org/10.1016/j.jbiomech.2016.12.027 Chandler RF, Clauser CE, McConville JT, Reynolds HM, Young JW (1975) Investigation of inertial properties of the human body. Aerospace Medical Research Laboratory, Wright-Patterson Air Force Base, Dayton Cheng C-K, Chen H-H, Chen C-S, Lee C-L, Chen C-Y (2000) Segment inertial properties of Chinese adults determined from magnetic resonance imaging. Clin Biomech 15(8):559–566. https://doi.org/10.1016/S0268-0033(00)00016-4 Clauser CE, McConville JT, Young JW (1969) Weight, volume, and center of mass of segments of the human body. Aerospace Medical Research Laboratory, Wright–Patterson Air Force Base, Dayton Dao TT, Marin F, Pouletaut P, Charleux F, Aufaure P, Ho Ba Tho MC (2012) Estimation of accuracy of patient-specific musculoskeletal modelling: case study on a post polio residual paralysis subject. Comput Methods Biomech Biomed Eng 15(7):745–751. https://doi.org/10.1080/ 10255842.2011.558086 Davidson PL, Wilson SJ, Wilson BD, Chalmers DJ (2008) Estimating subject-specific body segment parameters using a 3-dimensional modeller program. J Biomech 41(16):3506–3510. https://doi.org/10.1016/j.jbiomech.2008.09.021 Davis RB III, Õunpuu S, Tyburski D, Gage JR (1991) A gait analysis data collection and reduction technique. Hum Mov Sci 10(5):575–587. https://doi.org/10.1016/0167-9457(91)90046-Z de Leva P (1996a) Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J Biomech 29(9):1223–1230. https://doi.org/10.1016/0021-9290(95)00178-6 de Leva P (1996b) Joint center longitudinal positions computed from a selected subset of Chandler’s data. J Biomech 29(9):1231–1233. https://doi.org/10.1016/0021-9290(96)00021-8 Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, Guendelman E, Thelen DG (2007) OpenSim: Open-source software to create and analyze dynamic simulations of movement. IEEE Trans Biomed Eng 54(11):1940–1950. https://doi.org/10.1109/TBME.2007.901024 Dempster WT (1955) Space requirements for the seated operator. Wright Air Development Center, Wright-Patterson Air Force Base, Dayton
76
R. Dumas and J. Wojtusch
Drillis R, Contini R, Bluestein M (1964) Body segment parameters: a survey of measurement techniques. Artif Limbs 8(1):44–66 Dumas R, Aissaoui R, Mitton D, Skalli W, de Guise JA (2005) Personalized body segment parameters from biplanar low-dose radiography. IEEE Trans Biomed Eng 52(10):1756–1763. https://doi.org/10.1109/TBME.2005.855711 Dumas R, Cheze L, Verriest JP (2007a) Adjustments to McConville et al. and Young et al. body segment inertial parameters. J Biomech 40(3):543–553. https://doi.org/10.1016/j.jbiomech. 2006.02.013 Dumas R, Cheze L, Verriest JP (2007b) Corrigendum to “Adjustments to McConville et al. and Young et al. body segment inertial parameters”. J Biomech 40(7):1651–1652. https://doi.org/ 10.1016/j.jbiomech.2006.07.016 Dumas R, Robert T, Cheze L, Verriest J-P (2015) Thorax and abdomen body segment inertial parameters adjusted from McConville et al. and Young et al. Int Biomech 2(1):113–118. https:// doi.org/10.1080/23335432.2015.1112244 Durkin JL, Dowling JJ, Andrews DM (2002) The measurement of body segment inertial parameters using dual energy X-ray absorptiometry. J Biomech 35(12):1575–1580. https://doi.org/10.1016/ S0021-9290(02)00227-0 Ganley KJ, Powers CM (2004) Determination of lower extremity anthropometric parameters using dual energy X-ray absorptiometry: the influence on net joint moments during gait. Clin Biomech 19(1):50–56. https://doi.org/10.1016/j.clinbiomech.2003.08.002 Hinrichs RN (1990) Adjustments to the segment center of mass proportions of Clauser et al. (1969). J Biomech 23(9):949–951. https://doi.org/10.1016/0021-9290(90)90361-6 Ho Hoang K-L, Mombaur K (2015) Adjustments to de Leva-anthropometric regression data for the changes in body proportions in elderly humans. J Biomech 48(13):3732–3736. https://doi.org/ 10.1016/j.jbiomech.2015.08.018 Jackson JN, Hass CJ, Fregly BJ (2015) Residual elimination algorithm enhancements to improve foot motion tracking during forward dynamic simulations of Gait. J Biomech Eng 137(11):111002. https://doi.org/10.1115/1.4031418 Jensen RK (1978) Estimation of the biomechanical properties of three body types using a photogrammetric method. J Biomech 11(8-9):349–358. https://doi.org/10.1016/0021-9290(78) 90069-6 Jovic J, Escande A, Ayusawa K, Yoshida E, Kheddar A, Venture G (2016) Humanoid and human inertia parameter identification using hierarchical optimization. IEEE Trans Robot 32(3):726–735. https://doi.org/10.1109/TRO.2016.2558190 Kadaba MP, Ramakrishnan HK, Wootten ME (1990) Measurement of lower extremity kinematics during level walking. J Orthop Res 8(3):383–392. https://doi.org/10.1002/jor.1100080310 McConville JT, Churchill TD, Kaleps I, Clauser CE, Cuzzi J (1980) Anthropometric relationships of body and body segment moments of inertia. Aerospace Medical Research Laboratory, Wright-Patterson Air Force Base, Dayton Mungiole M, Martin PE (1990) Estimating segment inertial properties: comparison of magnetic resonance imaging with existing methods. J Biomech 23(10):1039–1046. https://doi.org/ 10.1016/0021-9290(90)90319-X Pearsall DJ, Reid G (1994) The study of human body segment parameters in biomechanics. Sports Med 18(2):126–140. https://doi.org/10.2165/00007256-199418020-00005 Pearsall DJ, Reid JG, Livingston LA (1996) Segmental inertial parameters of the human trunk as determined from computed tomography. Ann Biomed Eng 24(2):198–210. https://doi.org/ 10.1007/BF02667349 Pillet H, Bonnet X, Lavaste F, Skalli W (2010) Evaluation of force plate-less estimation of the trajectory of the centre of pressure during gait. Comparison of two anthropometric models. Gait Posture 31(2):147–152. https://doi.org/10.1016/j.gaitpost.2009.09.014 Plagenhoef S (1971) Patterns of human motion: a cinematographic analysis. Prentice-Hall, Englewood Cliffs
Estimation of the Body Segment Inertial Parameters for the Rigid Body. . .
77
Reid JG, Jensen RK (1990) Human body segment inertia parameters: a survey and status report. Exerc Sport Sci Rev 18(1) Sreenivasa M, Chamorro CJG, Gonzalez-Alvarado D, Rettig O, Wolf SI (2016) Patient-specific bone geometry and segment inertia from MRI images for model-based analysis of pathological gait. J Biomech 49(9):1918–1925. https://doi.org/10.1016/j.jbiomech.2016.05.001 Taddei F, Martelli S, Valente G, Leardini A, Benedetti MG, Manfrini M, Viceconti M (2012) Femoral loads during gait in a patient with massive skeletal reconstruction. Clin Biomech 27(3):273–280. https://doi.org/10.1016/j.clinbiomech.2011.09.006 Tisserand R, Robert T, Dumas R, Chèze L (2016) A simplified marker set to define the center of mass for stability analysis in dynamic situations. Gait Posture 48:64–67. https://doi.org/ 10.1016/j.gaitpost.2016.04.032 Valente G, Pitto L, Testi D, Seth A, Delp SL, Stagni R, Viceconti M, Taddei F (2014) Are subjectspecific musculoskeletal models robust to the uncertainties in parameter identification? PLoS One 9(11):e112625. https://doi.org/10.1371/journal.pone.0112625 Vaughan CL, Andrews JG, Hay JG (1982) Selection of body segment parameters by optimization methods. J Biomech Eng 104(1):38–44. https://doi.org/10.1115/1.3138301 Verriest JP (2012) Automatic anthropometric personalization of a digital human model from a set of subject’s photographs. Work 41(Suppl 1):4061–4068. https://doi.org/10.3233/WOR-20120071-4061 Winter DA (2009) Biomechanics and motor control of human movement. Wiley, New York Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D, Whittle M, D’Lima DD, Cristofolini L, Witte H, Schmid O, Stokes I (2002) ISB recommendation on definitions of joint coordinate system of various joints for the reporting of human joint motion – part I: ankle, hip, and spine. J Biomech 35(4):543–548. https://doi.org/10.1016/S0021-9290(01)00222-6 Wu G, van der Helm FCT, Veeger HEJ, Makhsous M, Van Roy P, Anglin C, Nagels J, Karduna AR, McQuade K, Wang X, Werner FW, Buchholz B (2005) ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion – part II: shoulder, elbow, wrist and hand. J Biomech 38(5):981–992. https://doi.org/10.1016/j.jbiomech. 2004.05.042 Yang F, Pai Y-C (2014) Can sacral marker approximate center of mass during gait and slip-fall recovery among community-dwelling older adults? J Biomech 47(16):3807–3812. https://doi. org/10.1016/j.jbiomech.2014.10.027 Yeadon MR, Morlock M (1989) The appropriate use of regression equations for the estimation of segmental inertia parameters. J Biomech 22(6):683–689. https://doi.org/10.1016/0021-9290 (89)90018-3 Young JW, Chandler RF, Snow CC, Robinette KM, Zehner GF, Lofberg MS (1983) Anthropometric and mass distribution characteristics of the adults female. FAA Civil Aeromedical Institute, Oklaoma City Zatsiorsky VM, Seluyanov VN, Chugunova LG (1990) Methods of determining mass-inertial characteristics of human body segments. In: Chernyi GG, Regirer SA (eds) Contemporary problems of biomechanics. CRC Press, Massachusetts, pp 272–291
Part II Discriminative Methods in Dynamic Pose Estimation
3D Dynamic Pose Estimation from Marker-Based Optical Data W. Scott Selbie and Marcus J. Brown
Abstract
The desire to capture images of human movement has existed since prehistoric times (see chapter “Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries”). However, it is only since the late nineteenth century and the development of cameras able to capture multiple sequential images that the recording and quantitative analysis of movement has become possible. With modern cameras and high computational power now available, it is commonplace for researchers and clinicians to make detailed measurements, from which an estimation of the position and orientation (pose) of a human body during motion can be computed. This chapter focuses on the estimation of dynamic 3D pose based on optical motion capture systems that record the 3D location of markers attached to the body (see Fig. 1). In this chapter, we describe the estimation of the pose of a multibody model comprising segments that are connected by joints that constrain the direction and range of motion between those segments. There are three common deterministic solutions to the problem of pose estimation; direct, single body, and multibody. This chapter focuses on the two optimization methods, single body and multibody, that provide a deterministic and a discriminative solution to the problem of pose estimation. Unlike the direct pose estimation, these two approaches mitigate, to some extent, uncertainty in the data.
W.S. Selbie (*) HAS-Motion, Inc., Kingston, ON, Canada C-Motion Inc., Germantown, MD, USA e-mail: [email protected] M.J. Brown HAS-Motion, Inc., Kingston, ON, Canada e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_152
81
82
W.S. Selbie and M.J. Brown
Keywords
Skeletal modeling • Pose estimation • Motion-capture • Inverse kinematics • Soft tissue artifact • Optimization
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Six Degree of Freedom (6DOF) Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pose Estimation Using a Technical Reference Frame (TF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inverse Kinematics (IK) Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Observability of the Inverse Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IK Optimization Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Direction Search Methods (Newton’s Method) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Global Search Methods: Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6DOF Versus IK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82 84 84 86 88 90 90 94 94 96 97 98 99 99
Introduction For this chapter, the assumption underlying pose estimation is that the human body model is constructed from a set of rigid (nondeformable) segments (or bodies) (see ▶ “Three-Dimensional Reconstruction of the Human Skeleton in Motion”). While this is not literally true, it allows a straightforward model that, for many biomechanical analyses, provides an adequate representation of the underlying skeletal structure for describing motor coordination and functional performance. Each segment is defined by a local anatomical reference frame (Cartesian coordinate system). These subjectspecific anatomical reference frames (AF) are often defined by the location of anatomically palpable landmarks, by matching statistical shape models to surface geometry, or by system identification methods such as functional joints estimated from recorded movements. Regardless of the technique used to establish the reference frame, the common goal is to establish an anatomically relevant reference frame that can be determined reliably and reproducibly. The origin of the reference frame can be located anywhere, but for convenience in this chapter the origin is placed at the proximal end of a segment coincident with the distal end of an adjacent segment (a joint connecting to the parent segment) (Fig. 1). Each segment is restricted to having one parent segment, and the segment’s interaction with its parent segment is described by the specification of joint constraints acting at and around the origin of a segment relative to the parent segment. These joint constraints define the number of degrees of freedom and possibly a prescribed relative path of the segments comprising the joint. The number of degrees of freedom can be any integer value between zero and six. A joint constrained to zero degrees of freedom with no path constraint allows no relative motion between segments, while a joint with six degrees of freedom allows the segments to move independently of each other.
3D Dynamic Pose Estimation from Marker-Based Optical Data
83
Fig. 1 A rigid body (part of a multibody model) is defined by an anatomical reference frame (AF) drawn as red vectors. The left leg is shown with N tracking markers mi and vectors ai describing the location of mi in AF for the right thigh and shank. Note that this configuration of tracking markers is but one of many different configurations in common use. The right leg displays a cluster of markers (sometimes secured to a rigid shell) and the left leg displays skin-based tracking markers attached at palpable anatomical landmarks
To estimate the pose of the multibody model, the 3D location of reflective markers attached to the segments is recorded by one or more optical sensors. It is beyond the scope of this chapter to describe the algorithms for identifying these 3D locations from the optical sensors, but regardless of the optical technology, the resultant 3D locations are used consistently between approaches. The tracking of each segment (pose estimation during a dynamic trial) is accomplished by establishing the location of the markers in the segment’s anatomical reference frame to which they are attached, recording the location of these markers in each frame of a motion trial, and by satisfying the specified joint constraints. A fundamental assumption of the algorithms presented in this chapter is that segments are rigid and the markers attached to those segments are secured rigidly and do not move relative to the segment to which they are attached. The number of markers required, and the number of segments to which markers are attached, depends on the structure of the multibody model and the pose estimation algorithm being used. The most important concept within this methodology is observability. Observability is dealt with in more detail later in the chapter; however, in short, a system is observable if the data are sufficient to describe, uniquely, the pose of the model. If the markers were truly attached rigidly to the underlying skeleton, i.e., a marker’s coordinates in the AF were invariant during movement, and the segments of the multibody model were truly rigid, and the markers were never occluded, this would be a straightforward chapter as all the pose estimation methods described in the scientific literature and textbooks would yield reliable pose estimations, and we could choose the mathematically simplest approach.
84
W.S. Selbie and M.J. Brown
Any marker that is attached to the skin, however, can move relative to the underlying skeleton (Cappozzo et al. 1996). This relative motion occurs as flesh between the marker and skeleton deforms during movements, and is commonly known in the biomechanics community as soft tissue artifact (STA). It is, as yet, challenging to mitigate STA through mathematical approaches because, while STA is systematic, it varies on a case-by-case basis between individuals, between locations on the body, and between movements. Pose estimation algorithms that mitigate these “uncertainties” resulting from STA can improve the effectiveness of pose estimation dramatically. The two pose estimation algorithms discussed in this chapter are common in the biomechanics community and are deterministic and discriminative. In other words, they rely solely on the structure of the multibody model and instantaneous data to estimate pose. This is in contrast with probabilistic pose estimation, in which prior information (e.g., models of STA or predictions based on the statistics of past performance) are incorporated into the pose estimation algorithm (see chapter ▶ “3D Dynamic Probabilistic Pose Estimation From Data Collected Using Cameras and Reflective Markers”).
State of the Art Six Degree of Freedom (6DOF) Pose Estimation This section describes an algorithm for six degree of freedom (6DOF) pose estimation, sometimes referred to as a segment optimization algorithm (Lu and O’Connor 1999) or single-body optimization. To estimate the pose of a segment at each frame of data, the 6DOF algorithm requires that a set of not less than three noncollinear markers be attached to each segment. To clarify the need for three markers, we will describe the information available from 1, 2, or more markers on a segment. If a segment was to have a single marker attached to it, this marker would permit the estimation of translations of the segment along the three principal axes of the global reference frame (e.g., 3DOF). If a second marker was added, it would be possible to estimate rotations about two principal axes of the segment; however, rotations about an axis between the two markers would be undetectable (e.g., 5DOF). When a third marker is added, offset from the line between the first two, rotations about all three segmental axes become observable (e.g., 6DOF). Additional markers on a segment cannot increase the number of degrees of freedom but, as will be see below, can be useful in a least-squares sense. This method is referred to as a 6DOF method because each segment (or joint) is considered to have six independent variables that describe its pose; three variables describe the location of the segment’s origin within the global reference frame (its position) and three variables describe the rotation about each of the principal axes of the segment (its orientation). In principle, each segment can be tracked independently of any other
3D Dynamic Pose Estimation from Marker-Based Optical Data
85
segment. This independence infers that there is no explicit linkage defined, i.e., there are no preconceived assumptions about the properties of any joint connecting segments. This means that the endpoints of a segment, and those of its the proximal and distal adjacent neighbors, are free to move relative to each other, based directly and solely on the recorded motion capture (MoCap) data (Cappozzo et al. 1995). This independent estimation of the pose of the segments requires that markers used to track one segment are not used to track any other segment. It is quite common, however, for one marker to be used as a tracking marker on two adjacent segments. For example, a lateral knee marker may be used as a tracking marker on the thigh and the shank. In this situation, the thigh and shank segments are still 6DOF because six variables describe the motion of a segment, but in this case the segments are not actually independent of each other. The 6DOF algorithm we describe here estimates the pose of a segment using a least-squares procedure (Kepple and Stanhope 2000). Consider a point mi attached to a segment, whose location is represented by vector ai in the AF. The location of the same marker mi is represented by vector vi in the GF (vi = the data recorded). The relationship between ai and vi is given by: vi ¼ RAG ai þ OAG
(1)
where: RAG is a rotation matrix from AF to GF OAG is the translation from AF to GF. The rotation matrix RAG and translation vector OAG may be computed at any instant, given that at least three noncollinear vectors ai are assumed stationary in the AF, and vi are recorded in GF, by minimizing the sum of squares error expression: f ðRAG , OAG Þ ¼
N X
ðvi RAG ai OAG Þ2
(2)
i¼1
where N is equal to the number of tracking targets on the segment. There are an infinite number of solutions of RAG and OAG that will produce minima for Eq. 2. Not all of these solutions result in RAG being a rotation matrix, so we specify the orthonormal constraint RTAG RAG ¼ I as a boundary condition on the solution (Spoor and Veldpaus 1980): gðRAG Þ ¼ RTAG RAG I ¼ 0
(3)
The method sets the gradient of Eq. 2 equal to the gradient of Eq. 3 times a set of Lagrangian multipliers: ∇f ðRAG , OAG Þ ¼ λ∇gðRAG Þ
86
W.S. Selbie and M.J. Brown
This results in a system of algebraic equations: ∇f ðRAG , OAG Þ λ∇gðRAG Þ ¼ 0
(4)
for which there exists an exact solution as long as N 3 The 6DOF algorithm requires a minimum of three noncollinear tracking markers, but more can be accommodated because the 6DOF algorithm permits a solution for an over-specified system with an unlimited number of tracking markers on a segment. This over-specification means that, provided noise (or some features of STA) in the data is uncorrelated, the least-squares algorithm will act to minimize the effects of the noise. If one or more tracking targets are missing in any frame(s), the over-specification still allows a calculated segmental pose, provided at least three noncollinear targets are present. The observability for a 6DOF method is straightforward because it is simply N 3, provided the locations of the markers are fixed in the AF, and are not collinear. In principle, tracking markers can be placed anywhere on a rigid segment. In practice, marker placement on an anatomical segment is a compromise between distributing markers over the entire surface of a segment and placing markers in areas that exhibit minimal STA (Cappozzo et al. 1997). As concluded in a review article by Cereatti et al. (2006), there have been attempts to modify the 6DOF algorithm in order to mitigate the effects of STA (Cappozzo et al. 1997; Andriacchi et al. 1998), but none of these approaches have proved satisfactory.
Pose Estimation Using a Technical Reference Frame (TF) While this chapter is focused on estimating pose from marker data, it is convenient at this time to discuss briefly pose estimation from two other 6DOF sensors: electromagnetic sensors and Moiré-phase tracking. It is beyond the scope of this chapter to describe the theory behind the sensor technology, but in summary, electromagnetic systems record the 6DOF pose of a sensor relative to an emitted electromagnetic dipole field. The Moiré-phase tracking (MPT) 3D motion capture system (Weinhandl et al. 2010) is a single-camera 3D motion tracking technology that tracks the 6DOF pose of a Moiré target (a lightweight, multilayer passive optical target; Weinhandl et al. 2010). The important idea to note is that these sensors describe their pose relative to an internal reference frame, not an anatomical frame. To put these 6DOF sensors in the context of marker-based MoCap (the focus of the chapter), we consider a slightly different approach to the 6DOF algorithm. Consider the same markers mi from Fig. 1, but instead of creating vectors ai in the anatomical reference frame of the segment, we create vectors bi in a technical reference frame (TF) defined by the markers (Fig. 2). In this description, the segment origin is located at one of the markers (m2), the principal axis is defined by vector from (m2 to m1), and the reference frame is established from the principal axis and m3. This adds another “layer” to the pose estimation as it requires an additional step to include the transformation between this TF and the associated AF (RTA, OTA).
3D Dynamic Pose Estimation from Marker-Based Optical Data
87
Fig. 2 A rigid body (part of a multibody model) is defined by an anatomical reference frame (AF) drawn as red vectors. The left leg is shown with N tracking markers mi and vectors bi describing the location of mi in a technical reference frame TF for the right thigh and shank. The left leg displays a Moiré-phase tracking sensor (top) and electromagnetic sensor (middle)
Using the same markers (mi) from one frame of data, and assuming that the transformation from TF to AF is invariant, we can identify (RTA, OTA) from vector calculus using the same methods used to define AF in the first place. Consider a point mi attached to a segment, whose location is represented by vector bi in the TF. Eq. (1) is written as: vi ¼ RTG bi þ OTG
(5)
where: RTG is a rotation matrix from TF to GF OTG is the translation from TF to GF RTG and OTG are computed as in Eq. 4. The resulting pose estimation (RTG, OTG) in a local reference frame, defined by the markers independently of the anatomy, is similar to the pose estimates of the other 6DOF sensors. In marker-based MoCap, it is possible to define the relationship
88
W.S. Selbie and M.J. Brown
between the markers and the anatomy because markers can be placed in locations that have anatomical meaning. With the electromagnetic and MPT technologies, the anatomical locations can be identified with a pointer, or system identification methods can be used to identify the AF. As in the previous section, our goal is once again to identify (RAG,OAG), but in this case, the least-squares solution computes the transformation from the TF to the GF (RTG,OTG). The additional step is to include the additional transform from TF to AF (RTA,OTA). RAG ¼ RTTA RTG OAG ¼ OTG OAT
(6)
There is a considerable benefit to the 6DOF approach to pose estimation, as it is straightforward to implement with results that are easy to understand. The 6DOF solution has no local minima, and requires no guidance from users. Notably, 6DOF estimates a pose that is an accurate representation of the data, which is useful for identifying local problems. An example of such a local problem would be the swapping of the names of two markers between trials, or even within a trial (something not uncommon when working with many passive reflected markerbased MoCap systems). Such mislabeling of markers will cause obvious discontinuities in the pose estimations of a 6DOF segment, which can be easily identified and corrected. The deterministic assumption that neither STA nor noisy marker data occur can result in pose estimations where the adjacent endpoints of segments are dislocated from each other or “merge” together. While these pose solutions reflect the true marker data, and thus highlight the presence of noise and/or STA, they can present estimations of pose that are anatomically impossible. To highlight the serious challenge of STA, if the entire set of markers translates in unison (e.g., through inertial forces or impact), the estimated pose of the segment can be quite wrong. There is, however, no information in the relative configuration of the tracking markers to indicate that anything has gone awry, so this artifact cannot be mitigated. The next section describing inverse kinematics discusses a deterministic approach to remove such an obvious artifact as joint disarticulation from the 6DOF model.
Inverse Kinematics (IK) Pose Estimation Inverse kinematics (IK) is the search for, and identification of, an optimal pose of a multibody model with explicit joint constraints, such that the overall differences between the measured and model-estimated marker coordinates are minimized, in a least-squares sense, at a system level. Lu and O’Connor (1999) termed this process global optimization, but in this chapter, we will refer to this as multibody optimization or IK. IK, as described here, is a least-squares solution that may be considered an extension to the 6DOF pose estimation because if a joint is ascribed six degrees of
3D Dynamic Pose Estimation from Marker-Based Optical Data
89
Fig. 3 A multibody model showing the pelvis as a root segment (e.g., 6DOF with respect to the global reference frame) and joint constraints at the hip, knee, and ankle of the left leg. In this figure, the hip has 3DOF, the knee 5DOF, and the ankle 3DOF, but many other multibody configurations can be found in the literature
freedom within the IK, the IK and 6DOF solutions are equivalent. Selecting appropriate joint constraints is idiosyncratically based on the number of markers being tracked, the context of the motion being analyzed, and many other factors; some of these factors will be discussed later in this section. As with 6DOF, the algorithms involved in IK pose estimation will be described in the context of marker-based optical 3D MoCap (Fig. 3). The solution to the IK is the pose of a multibody model that best matches the MoCap data, in terms of a least-squares criterion. In the Lu and O’Connor (1999) approach, the IK solution is found for each frame of data, independent of any previous or subsequent frames of data. Mathematically, van den Bogert and Su (2008) described this approach, based on the overall configuration of the multibody model, using a set of generalized coordinates q.
90
W.S. Selbie and M.J. Brown
Generalized coordinates are the minimum set of independent variables that describe the pose. In this case, R and O of Eq. 1 now consist of multiple transformations and become a function of the generalized coordinate vector q: vi ¼ RðqÞai þ OðqÞ
(7)
The expression that is minimized becomes: f ðR, OÞ ¼ f ðqÞ ¼
N X
ðvi RðqÞai OðqÞÞ2
(8)
i¼1
where N is the total number of targets on all the segments in the IK chain.
Weighting Within an IK model, it is possible to rely more on data that are known, a priori, to contain less noise or be less affected by STA. This can be achieved via a weighting. f ðR, OÞ ¼ f ðqÞ ¼
N X
αi ðvi RðqÞai OðqÞÞ2
(9)
i¼1
The selection of the weights, αi, can be made pragmatically and heuristically, or rules may be used that allow the computation of an optimal set of weights. Without a priori information, it is usually best to set αi to 1, but on occasion, when estimating pose, the user may want to ensure that certain segments follow the tracking targets with a higher degree of accuracy than other segments. For example, the user may want the distance between the foot and the floor (or recorded ground reaction force) to remain similar to the values that would be obtained using a 6DOF method because 6DOF is likely the best local estimate of the pose of the foot. Likewise, data from some markers may not be considered representative of the pose because they are noisy, so the weight of these data can be reduced. In some cases, the marker may be known to have substantial STA relative to one of the degrees of freedom (generalized coordinates) and the influence of the marker on this generalized coordinate can be removed.
Observability of the Inverse Kinematics As mentioned previously, the pose of a multibody model is observable if the data are sufficient to describe the pose uniquely. In the case of the 6DOF pose estimation, three or more rigidly attached, noncollinear targets are required to track each segment. When one target is placed on a rigid segment, three independent pieces of information can be obtained, the X, Y, and Z coordinates of the target. When a
3D Dynamic Pose Estimation from Marker-Based Optical Data
91
second target is placed on the segment, two further pieces of information are obtained. The number of new pieces of information for the second target is two, not three like the first target, because if we know the X and Y locations of the second target, then the Z coordinate is known because the distance between the first and second targets is fixed. Thus, two targets only supply five of the six unknowns. When a third target is added, one additional piece of information is supplied; note the third target only adds one new piece of information because the distance from the third target to the first target and the distance from the third target to the second target are fixed. Still with three noncollinear targets, we have sufficient information to fully solve the pose of a 6DOF segment. With IK, not only is there the assumption of rigid segments, but there are also constraints added at the joints. A consequence of the joint constraints is that fewer than three markers may be sufficient to fully determine the pose of a segment. For example, a segment that has only one degree of freedom (e.g., one connected to a parent segment by a hinge joint) only requires one marker to fully determine the joint angle. It is not possible to just count markers, however, because if this one marker is coincident with the hinge joint, it does not provide any information and the pose is nonobservable. Therefore, the question of whether the markers provide sufficient information to determine the model’s pose is far more complex when joint constraints exist. A straightforward approach to the problem would be to specify the number of targets required to track a segment, based solely on the type of joint connecting that segment to its parent. For example, Yeadon (1984) required two markers to track a segment connected to the parent via a ball joint (three degrees of freedom) or a universal joint (two degrees of freedom) and required only one marker when the segment was connected via a one degree of freedom hinge joint. Although this approach will guarantee that the system will likely be observable, if these requirements are met, it can be overly conservative and will occasionally consider the model to be unobservable, when in fact there is sufficient information available. For example, Schulz and Kimmel (2010) demonstrated that it is possible to track the pose of the thigh segment without actually placing any markers on the thigh. Yeadon’s method would declare this model to be unobservable. This is important because for many activities, the STA of markers on the thigh is detrimental to an accurate estimate of the pose and if Schulz’s assumption that the hip has three degrees of freedom and the knee has one degree of freedom is an accurate reflection of the movement, his approach could be useful for studying many activities. To demonstrate how it is possible to calculate a general solution to the observability problem, consider the simple example of a single segment constrained to its parent (in this example, the ground) by a ball joint. This system can be fully described by three degrees of freedom: the Euler rotations, θx , θy , and θz. For this case, the general IK objective function Eq. 8 becomes: P ð qÞ ¼
m X i¼1
fðR0 ðqÞAi Þg
2
(10)
92
W.S. Selbie and M.J. Brown
Assume there is only one target (m = 1) fixed to the segment, the local coordinates, in the AF, of that target (Ax, Ay, Az) and the global coordinates of the targets, in the GF, are (Px, Py, Pz). Applying Eq. (10) for this simple case of one segment connected to the ground via a ball joint with a single tracking target, the objective function f(q) is: Pðq Þ AX cos ðθz Þ cos θy þ Ay cos ðθz Þ sin θy sin ðθx Þ þ sin ðθz Þ cos ðθx Þ þ Az cos ðθz Þ sin θy cos ðθx Þ þ sin ðθz Þ sin ðθx Þ Px ¼ Ax sin ðθz Þ cos θy þ Ay sin ðθz Þ sin θy sin ðθx Þ þ cosðθz Þ cos ðθx Þ þ Az sin ð θ Þ sin θ ð Þ þ cos ð θ Þ sin ð θ Þ P cos θ y x z x y z AX sin θy þ Ay cos θy sin ðθx Þ þ Az cos θy cos ðθx Þ Pz
If a change in rotation (some combination of a change in θx , θy , θz) exists for which the target does not move, then the system is not observable. In order to establish whether this is the case, it is necessary to discover if a situation exists where the cost function does not change with respect to changes in the joint angle. This exactly describes the Jacobian (or matrix of partial derivatives) of the cost function: d ð Px Þ d ð Px Þ d ð Px Þ d ðθx Þ d θy dðθz Þ d P d Py d Py y Jacobian ¼ d ðθx Þ d θy d ðθ z Þ d ð Pz Þ d ð Pz Þ d ð Pz Þ d ðθx Þ d θy d ðθ z Þ Calculating the Jacobian of the cost function described in Eq. 10: d ðP x Þ ¼ Ay cos ðθz Þ sin θy cos ðθx Þ sin ðθz Þ sin ðθx Þ d ðθ x Þ þ Az cos ðθz Þ sin θy sin ðθx Þ þ sin ðθz Þ cos ðθx Þ d ð Px Þ ¼ Ax cos ðθz Þ sin θy þ Ay sin ðθz Þ cos θy sin ðθx Þ sin ðθz Þ sin ðθx Þ d θy þ Az cos ðθz Þ cos θy cos ðθx Þ d ð Px Þ ¼ Ax sin ðθz Þ cos θy þ Ay sin ðθz Þ sin θy sin θx þ cos ðθz Þ cos ðθx Þ d ðθ z Þ þ Az sin ðθz Þ sin θy sin ðθx Þ þ cos ðθz Þ sin ðθx Þ d Py ¼ Ay sin ðθz Þ sin θy cos ðθx Þ cos ðθz Þ sin ðθx Þ d ðθ x Þ þ Az sin ðθz Þ sin θy sin ðθx Þ þ cos ðθz Þ cos ðθx Þ d Py ¼ Ax sin ðθz Þ sin θy þ Ay sin ðθz Þ cos θy sin ðθx Þ cos ðθz Þ sin ðθx Þ d θy þ Az sin ðθz Þ cos θy cos ðθx Þ
3D Dynamic Pose Estimation from Marker-Based Optical Data
93
d Py ¼ Ax cos ðθz Þ cos θy þ Ay cos ðθz Þ sin θy sin ðθx Þ sin ðθz Þ cos θx d ðθz Þ þ Az cos ðθz Þ sin θy sin ðθx Þ sin ðθz Þ sin ðθx Þ d ð Pz Þ ¼ Ay cos θy cos ðθx Þ Az cos θy sin ðθx Þ d ðθ x Þ d ð Pz Þ ¼ Ax cos θy þ Ay sin θy sin ðθx Þ Az sin θy cos ðθx Þ d θy d ð Pz Þ ¼0 d ðθ z Þ To simplify this equation, consider the state where θx = 0, θy = 0, The Jacobian now becomes: 0 Az Ay 0 Ax Jacobian of cost function ¼ Az Ay Ax 0
θz = 0
The determinant of the Jacobian is: Det ¼ 0ð0 Ax Ax Þ AZ Ax Ay 0 þ Ay ð Az Ax 0Þ ¼ 0 Ax Ay Az þ Ax Ay Az Since the determinant of the Jacobian is zero, it is not invertible and its rank is not full; thus, one target is not sufficient to estimate the pose of a segment connected to ground via a ball joint. Assume now that two targets are attached to the segment: A1 and A2. In this case, the cost function (Eq. 10) for one segment that is connected to ground by a ball joint is: Pðp Þ Ax1 cos ðθz Þ cos θy þ Ay cos ðθz Þ sin θy sin ðθx Þ þ sin ðθz Þ cos ðθx Þ þ Az1 cos ðθz Þ sin θy cos ðθx Þ þ sin ðθz Þ sin ðθx Þ Px1 1 Ax1 sin ðθz Þ cos θy þ Ay sin ðθz Þ sin θy sin ðθx Þ þ cos ðθz Þ cos ðθx Þ þ Az1 sin ðθz Þ sin θy cos ðθx Þ þ cos ðθz Þ sin ðθx Þ Py 1 þ A sin θ cos θ A sin θ cos θ ð Þ þ A cos θ ð Þ P x y y y x z y x z 1 1 1 ¼ Ax2 cos ðθz Þ cos θy þ Ay2 cos ðθz Þ sin θy sin ðθx Þ þ sin ðθz Þ cos ðθx Þ þ Az2 cos ðθz Þ sin θy cos ðθx Þ þ sin ðθz Þ sin ðθx Þ Px2 Ax sin ðθz Þ cos θy þ Ay sin ðθz Þ sin θy sin ðθx Þ þ cos ðθz Þ cos ðθx Þ þ Az sin ðθz Þ sin θy cos ðθx Þ þ cos ðθz Þ sin ðθx Þ Py 2 2 2 2 Ax2 sin θy þ Ay2 cos θy sin ðθx Þ þ Az2 cos θy cos ðθx Þ Pz2
Again, taking the simplest case and setting the orientation to θx ¼ 0, θy ¼ 0, θz ¼ 0 The Jacobian of the cost function now reduces to: 0 A1z A1y A1 z 0 A1x A1y A1x 0 Jacobian of cost function = A2z A2y 0 A2 z 0 A2x A2y A2x 0
94
W.S. Selbie and M.J. Brown
If targets A1 and A2 have coordinates (0, 0, A1z) and (0, 0, A2z) which they are collinear along the Z axis, we would expect the system to be unobservable as the targets will not register rotation about the Z axis. For this case: 0 A1z 0 0 0 A1z 0 0 0 Jacobian of cost function = 0 A2z 0 0 0 A2z 0 0 0 Column 3 equals zero, not full column rank, and thus the system is not observable as expected. Now If the two targets do not form a line that points to the joint center, for example: A1 ¼ ð0:1, 0:1, 0:1Þ and A2 ¼ ð0:1, 0:1, 0:1Þ The Jacobian now is: 0 0:1 0:1 Jacobian of cost function ¼ 0 0:1 0:1
0:1 0:1 0 0:1 0:1 0 0:1 0:1 0 0:1 0:1 0
This matrix has a rank = 3, which is full column rank and thus marker information (A1 and A2) is independent and the model is fully observable. Therefore, the general solution for observability in inverse kinematics reduces to determining whether the Jacobian for cost function of Eq. 10 has full rank. If it does, we have sufficient information to determine the pose of the model. Conversely, if the rank of the Jacobian of the IK cost function is not full rank, there is not enough information to determine a unique pose for the model.
IK Optimization Algorithms In the general case, there is no analytic solution for the IK problem. We, therefore, summarize examples from two classes of implementation of a numerical solution to this optimization problem: direction search methods and global search methods.
Direction Search Methods (Newton’s Method) To understand Newton’s method, consider a function f(q) that starts at an initial vector q0, moves through a series of vectors qk, and converges to a solution at qmin.
3D Dynamic Pose Estimation from Marker-Based Optical Data
95
Newton’s method is a three step process: 1. Compute the search direction 2. Determine the length of the next step 3. Use the results of steps 1 and 2 to obtain a new point qk. These steps are repeated until a minimum is found To find the search direction (Step 1) using Newton’s method, consider a vector, q, on f(q) located near the current value qk .The vector q qk can be approximated by a second-order Taylor series expansion: f ðq qk Þ ¼ f ðqk Þ þ Δf ðqk Þðq qk Þ þ
Bðqk Þðq qk Þ2 þ ... 2
(11)
where Δf(qk) is the gradient of f at the current value qk and B is the Hessian, or matrix, of second partial derivatives at qk. Taking the derivative of the function in Eq. 11 with respect to q and ignoring the derivative of the third term (e.g., the Hessian), we obtain: Δf ðq qk Þ ¼ Δf ðqk Þ þ Bðqk Þðq qk Þ
(12)
The derivative has a minimum at: 0 ¼ Δf ðqk Þ þ Bðqk Þðq qk Þ
(13)
and thus the search direction, (q qk), can be obtained from: ðq qk Þ ¼ Δf ðqk ÞB1 ðqk Þ
(14)
After solving for the search direction, (q qk), the next point in the search, qk + 1, is found by moving in the direction of (q qk). Ideally, the step size is determined by the magnitude of the eigenvalues of the movement to ensure that we obtain a sufficient decrease in the cost function, without taking excessively small steps. In practice, steps sizes that have worked for previous data sets are assumed to be sufficient. Once qk + 1 is obtained, it is checked against a termination criterion (is (q qk) small). If the termination criterion is satisfied, then the minimum for the global IK problem is found. If the criteria is not met, the process is repeated, beginning at step 1 with qk + 1 acting as the new current value qk. Ideally, Δf(qk) and B (Hessian) are derived symbolically but this is not always straightforward. Furthermore, even if the symbolic version of the Hessian is derived, computing the inverse of the Hessian, B1, requires a series of linear equations to be solved, which can be computationally costly. An alternative to this method, called the quasi-Newton method, Δf(qk) and B are approximated numerically by the change in the gradient between steps. Several methods of approximation have been proposed that all follow three primary assumptions:
96
W.S. Selbie and M.J. Brown
1. The Hessian must be symmetrical 2. The model gradient must be equal to the function gradient at the current step and at the previous step 3. The Hessian cannot change drastically between successive steps The consequence of these assumptions is that convergence may be compromised. Unlike the 6DOF least-squares solution, there are many possible solutions to the IK optimization as the solution space typically has many local minima. If the initial estimated position, q0, is “close” to the global minimum, the solution will likely converge to the correct solution. The initial estimated position, or “seed,” is therefore critical to the success of the algorithm. For the first frame of data, it is possible to use a 6DOF solution as the seed. For subsequent frames, the seed for the optimization algorithm at any given frame is the state of the model at the previous frame. This could be problematic if the solution at the previous frame was an inappropriate local minimum, resulting in subsequent pose estimates diverging from the real solution due to being held in this local minimum. For example, the data collection volumes of most optical MoCap systems are smaller than the laboratory that they are in, and subjects often begin their movements outside the volume (for example, to perhaps ensure that they are at a constant speed while walking or running through the data collection volume). The first frame with complete data can often be relatively unreliable because it is captured near to the edge of the calibrated volume, and therefore the likelihood of the optimization solution becoming trapped in a local minimum increases. In order to avoid this, one potential improvement to the algorithm is to compute the solution both forward and backward, in the hope that one of the passes will provide a more optimal solution path.
Global Search Methods: Simulated Annealing Simulated annealing (Higginson et al. 2005; Ingber 2012) is a Monte Carlo method in which the solution space is explored probabilistically by randomly searching near the best known solution. Simulated annealing is not prone to finding a local minima and therefore, given “enough” computing time (unfortunately, “enough” cannot be calculated but needs to be learned from experience), finds the global minimum. It is modeled after annealing in metallurgy, in which the thermodynamic free energy of a metal decreases as its temperature cools. In simulated annealing, as the virtual temperature cools, the algorithm searches in a smaller and smaller region around the best known solution (Fig. 4). Simulated annealing functions using two nonobvious principles: 1. Some new values that do not actually reduce the minimum value are allowed so that more of the solution space can be explored. (The allowed values are determined by the Metropolis criteria.) 2. After making many estimates, and observing that the cost function declines slowly, one lowers the temperature and thus limits the size of allowed values
3D Dynamic Pose Estimation from Marker-Based Optical Data
97
Fig. 4 Flowchart of the simulated annealing algorithm. The size of the perturbation is based on current temperature and the Metropolis criteria: Randð0, 1Þ < e
f i f best T
that are larger than the current minimum. After lowering the temperature several times, only more optimal values are accepted, and the optimization approaches the global minimum. One of the biggest challenges to simulated annealing is that the algorithm is computationally expensive, and perhaps more problematically, it is not possible to determine if the current solution is actually a global minimum without continuing the optimization indefinitely. In other words, there is no threshold or criterion for identifying that the search is complete. The user must decide how many iterations to perform in the optimization and accept that the minimum found in that time period may not be the global minimum. Despite the computational cost (time), simulated annealing is a more robust algorithm than direction search algorithms. Despite the robustness, however, most IK users opt for direction search algorithms because of time constraints.
6DOF Versus IK In many circumstances, the IK solution is likely to be more anatomically congruent and therefore preferable to the 6DOF solution, but the user must attend to the
98
W.S. Selbie and M.J. Brown
determination of the appropriateness of the selected joint constraints. For example, an experiment that was focused on understanding the kinematics of an injured knee, where translations and rotations occur as a result of the injury (e.g., anterior cruciate ligament damage), would likely not benefit from an IK solution where the constraints, and consequent prescribed motion of the knee joint, “hide” the pathology. Finally, it is well known that residual errors, i.e., differences between model predictions and marker measurements, computed by IK algorithms are reflections of noise in the marker data, soft tissue artifact, and inaccurate marker placement. A limitation of the IK algorithm, however, is that it has no straightforward mechanism to compensate for systematic noise, even though it can be used to identify its presence.
Future Directions In this chapter, we have described the current state of deterministic pose estimation algorithms. The future evolution of deterministic algorithms is quite limited. Begon et al. (2016), for example, has introduced an approach that removes STA without modeling STA but rather by ignoring information in markers that are considered unreliable. For many segments of the human body, STA has a particularly disastrous effect on the axial rotation of the segment. In other words, the markers rotate about the long axis of the segment (upper arm, forearm, thigh to name a few). Begon’s solution was to ignore any information in the marker that would reflect axial rotation by projecting tracking markers onto the long axis of the segment. These projected markers influence five of the degrees of freedom of a segment only. The long axis rotation is then estimated based on the pose of adjacent-constrained segment. The example given by Begon is movement of the upper arm, in which the axial rotation of the upper arm is estimated by constraining the elbow joint to have only two rotational degrees of freedom, and therefore the axial rotation of the upper arm is based on the pose of the forearm. There is some potential for improvements to deterministic pose estimation algorithms based on similarly clever rejection of data in isolated/idiosyncratic cases. It is our believe that the future of marker-based pose estimation lies not in deterministic algorithms but in algorithms based on Bayesian Inference (Todorov 2007) (chapter ▶ “3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers”) and algorithms based on optimal control theory (Miller and Hamill 2015) (▶ “Optimal Control Modeling of Human Movement”). Bayesian Inference allows a principled way to mitigate the effects of STA by modeling artifact and removing it. Optimal control theory is capable of generating motion independently of any recorded data based on generated simulated motion of the behavior based on some optimization criteria (e.g., minimum energy). The technique can be influenced by recorded data to ensure that the pose estimation is arbitrarily close to the recorded motion. Optimal control theory has the additional
3D Dynamic Pose Estimation from Marker-Based Optical Data
99
benefit of being able to generate solutions for unobservable, and even sparse, marker sets. Lastly, it is important to consider algorithms for which the soft tissue artifact is considered important data reflective of an individual subject instead of an artifact to be removed. Michael Black’s laboratory at the Max Planck Institute for Intelligent systems has been developing pose estimation algorithms based on statistical shape models (Loper et al. 2015). Instead of defining pose based on the position and orientation of an underlying skeleton, this research has focused on modeling the surface geometry of the subject and estimating the pose of the surface. Based on high-density surface scans of subjects performing movement, the statistical shape model is a parameterized surface that can be subsequently fit to sparse surface data (e.g., markers). These models are remarkably good at representing the surface of the body during motion. From a biomechanics perspective, a fundamental question is whether we can infer the multibody skeletal pose from this parameterized surface data.
Cross-References ▶ 3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers ▶ Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries ▶ Optimal Control Modeling of Human Movement ▶ Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units ▶ Three-Dimensional Reconstruction of the Human Skeleton in Motion
References Andriacchi TP, Alexander EJ, Toney MK, Dyrby C, Sum J (1998) A point cluster method for in vivo motion analysis: applied to a study of knee kinematics. J Biomech Eng 120:743–749 Begon M, Bélaise C, Naaim A, Lundberg A, Chèze L (2016) Multibody kinematics optimization with marker projection improves the accuracy of the humerus rotational kinematics. J Biomech (16):31111–31113 Cappozzo A, Catani F, Croce UD, Leardini A (1995) Position and orientation in space of bones during movement: anatomical definition and determination. Clin Biomech 10(4):171–178 Cappozzo A, Catani F, Leardini A, Benedetti MG, Della Croce U (1996) Position and orientation in space of bones during movement: experimental artefacts. Clin Biomech 11(2):90–100 Cappozzo A, Cappello A, Della Croce U, Pensalfini F (1997) Surface-marker cluster design criteria for 3-D bone movement reconstruction. IEEE Trans Biomed Eng 44(12):1165–1174 Cereatti A, Della Croce U, Cappozzo A (2006) Reconstruction of skeletal movement using skin markers: comparative assessment of bone pose estimators. J Neuro Eng Rehabil 3(1):7 Higginson JS, Neptune RR, Anderson FC (2005) Simulated parallel annealing within a neighborhood for optimization of biomechanical systems. J Biomech 38:1938–1942
100
W.S. Selbie and M.J. Brown
Ingber L (2012) In: Oliveira H, Petraglia A, Ingber L, Machado M, Petraglia M (eds) Adaptive simulated annealing, in stochastic global optimization and its applications with fuzzy adaptive simulated annealing. Springer, New York, pp 33–61 Kepple T, Stanhope S (2000) Moved software. In: Winters, Crago (eds) Biomechanics and neural control of posture and movement. Springer, New York Loper M, Mahmood N, Romero J, Pons-Mol G, Black MJ (2015) SMPL: a skinned multi-person linear model. ACM Trans Graph 34(6):248:1–248:16. ACM Lu TW, O’Connor JJ (1999) Bone position estimation from skin marker co-ordinates using global optimization with joint constraints. J Biomech 32:129–134 Miller R, Hamill J (2015) Optimal footfall patterns for cost minimization in running. J Biomech 48:2858–2864 Schulz BW, Kimmel WL (2010) Can hip and knee kinematics be improved by eliminating thigh markers?Clinical. Biomechanics 25(2010):687–692 Spoor C, Veldpaus F (1980) Rigid body motion calculated from spatial coordinates of markers. J Biomech 13(4):391–393 Todorov E (2007) Probabilistic inference of multijoint movements, skeletal parameters and marker attachment from diverse motion capture data. IEEE Trans on Biomed Eng 54:1927–1939 Van Den Bogert AJ, Su A (2008) A weighted least squares method for inverse dynamic analysis. Comput Methods Biomech Biomed Eng 11(1):3–9 Weinhandl JT, Armstrong BSR, Kusik TP, Barrows RT, O’Connor KM (2010) Validation of a single camera three-dimensional motion tracking system. J Biomech 43(7):1437–1440 Yeadon MR (1984) The mechanics of twisting somersaults. Doctoral thesis. University of Calgary
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography Hans Gray, Shanyuanye Guan, Peter Loan, and Marcus Pandy
Abstract
Accurate measurement of in vivo joint kinematics is important for understanding normal and pathological human motion and for evaluating the outcome of surgical procedures. Biplane videoradiography is currently the most accurate method available for measuring in vivo joint kinematics noninvasively. The method uses two X-ray images obtained from different perspectives to deduce precise three-dimensional spatial information of the bones that meet at a joint. The abilities to collect high-quality X-ray images at high frame rates and to process these images in a time efficient manner are key factors determining the feasibility of using modern biplane videoradiography systems to measure human joint motion in vivo. The latest developments in this field include improvements in image quality, software for more efficient and accurate data processing, and the advent of mobile biplane videoradiography systems. Mobile systems enable data capture for a wider range of joints and activities by increasing the effective image capture volume, thereby addressing a major limitation of stationary systems. This chapter summarizes the most recent advances in human motion measurement using biplane videoradiography (also commonly referred to as biplane X-ray fluoroscopy). We begin with some basic considerations related to hardware setup, data capture, and data processing and then describe methods commonly used to evaluate system accuracy. The chapter concludes with a discussion of the relative merits of mobile versus stationary systems as well as some thoughts on potential future applications of biplane videoradiography in human joint motion measurement. H. Gray (*) • S. Guan • M. Pandy Department of Mechanical Engineering, The University of Melbourne, Parkville, VIC, Australia e-mail: [email protected]; [email protected]; [email protected] P. Loan C-Motion, Inc., Germantown, MD, USA e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_154
101
102
H. Gray et al.
Keywords
3D joint kinematics • Pose estimation • Six-degree-of-freedom joint motion • Biplane fluoroscopy • Mobile biplane fluoroscopy
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hardware Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X-Ray Unit Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mobility of the X-Ray Imaging Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X-Ray Generator Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Frame Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exposure Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Motion Blur . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Image Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Validation of Measurement Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contrasting Mobile and Stationary Biplane Videoradiography Systems . . . . . . . . . . . . . . . . . . . . . . Conclusion and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
102 103 104 104 105 105 106 106 107 107 107 107 108 108 111 112 113 113
Introduction Accurate measurement of in vivo joint kinematics is important for understanding normal and pathological human motion and for evaluating the outcome of surgical procedures (Dennis et al. 2005). Biplane videoradiography is currently the most accurate method available for measuring in vivo joint kinematics noninvasively. Two X-ray video imaging units are used to acquire time-synchronized sequences of two-dimensional (2D) X-ray images of a target joint from two different orientations (Fig. 1). Threedimensional (3D) geometric models of the individual bones or implant components comprising the joint are then registered to the 2D images in a 3D computer space through a process called “pose estimation.” Pose estimation defines the position and orientation of each of the bones or implants in a common reference frame and is used to calculate joint kinematics. Joint kinematics describes the relative positions and orientations of the bones using anatomically meaningful parameters (Grood and Suntay 1983). The aim of this chapter is to summarize recent advances in human motion measurement using biplane videoradiography (a technique that is also commonly referred to as biplane fluoroscopy). The next section titled “State of the Art” briefly summarizes the history and current state of biplane videoradiography. In the subsequent three sections, we follow a process flow-based structure focusing on issues related to the configuration of biplane systems, data collection, and data processing (see Fig. 2). Section “Validation of Measurement Accuracy” addresses matters concerning the accuracy with which biplane videoradiography systems are able to measure joint kinematics. Section “Contrasting Mobile and Stationary Biplane Videoradiography Systems” compares
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
103
Fig. 1 Design of a typical biplane videoradiography system. The X-ray tubes generate X-ray beams which pass through the joint of interest and enter the image intensifiers. Each image intensifier projects onto a phosphor screen an image that is subsequently captured by a camera mounted on the intensifier. The two cameras collect the images in a time-synchronized manner
Fig. 2 Schematic diagram illustrating the key steps involved with using biplane videoradiography to measure joint kinematics
and contrasts mobile and stationary biplane systems, while Section “Conclusion and Future Directions” concludes with some thoughts on the future of biplane videoradiography with respect to its potential applications in human motion measurement.
State of the Art The use of X-ray images to investigate the relative positions of bones, fractured segments of a bone, or foreign bodies dates back to the late nineteenth century following the discovery of X-ray imaging. The use of two X-ray images taken from two perspectives for precise measurement of the relative position of human tissue and foreign bodies emerged soon thereafter (Davidson 1898). Biplane X-ray images have
104
H. Gray et al.
been used to estimate 3D joint kinematics at least as early as the 1970s. These early studies were performed under static conditions and used either discrete bony landmarks (Matteri et al. 1976) or tantalum beads inserted in the bone (Selvik 1974; van Dijk et al. 1979) to calculate the pose (position and orientation) of the bone. The technique of using tantalum beads embedded in bones to precisely measure skeletal kinematics is called roentgen stereophotogrammetric analysis (RSA) and was introduced by Selvik (1974). Several early studies used this technique to measure joint kinematics under static conditions and slow dynamic conditions (Kärrholm et al. 1988; Uvehammer et al. 2000). More recent studies have used the technique to study more rapid dynamic movements (Anderst et al. 2009; Tashman et al. 2007). Although radiopaque markerbased pose estimation is still widely used (Anderst et al. 2008; Brainerd et al. 2010), its invasive nature makes it unsuitable for clinical studies involving large numbers of patients. In order to overcome limitations imposed by the need for beads, markerless techniques were developed to estimate pose from X-ray images for joint replacements (Banks and Hodge 1996; List et al. 2012) and for intact bone (Giphart et al. 2012; You et al. 2001; Zhu and Li 2011). Over the last two decades, refinements in sequential X-ray image acquisition systems have enabled higher quality images (with higher resolution and better contrast) acquired at higher frame rates. Improvements in pose estimation software and increased computing power at reduced costs have further expanded the use of biplane videoradiography systems in measuring 3D joint kinematics in vivo.
Hardware Configuration X-Ray Unit Configuration A videoradiography unit consists of an X-ray source and an image collector usually referred to as an image intensifier. The image capture volume of a single X-ray unit forms a cone with the X-ray source at the apex and the image intensifier forming the base. Because biplane videoradiography requires simultaneous imaging of a joint using two X-ray units, it is necessary for the joint to be within the intersection of the image capture volumes of both X-ray units during the activity of interest. There are several factors to consider when deciding on the configuration of the X-ray units, including anatomical direction used in imaging the joint, occlusion, and the size and shape of the capture volume. Anatomical Direction of Imaging and Occlusion: The anatomical direction of imaging has a bearing on the image quality. For example, anterior-posterior (AP) views may give clearer images of the hip joint than medial-lateral views due to reduced depth of bone and soft tissue in the AP direction. For the tibiofemoral joint, equally clear images may be obtained from the AP as well as the medial-lateral (ML) direction, whereas for the patellofemoral joint, the ML direction is preferred as it reduces occlusion by the femur. However, ML images of the knee will be occluded by the contralateral leg during walking (Fig. 3). Capture volume: The size, shape, and orientation of the image capture volume are critical as the joint must remain within the capture volume, which is relatively small.
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
105
Fig. 3 Samples of a series of biplane (Plane 1 and Plane 2) X-ray images of the knee of a total knee replacement patient collected during overground walking. Images taken at around 20% and 80% of the gait cycle were occluded by the contralateral leg for Plane 1. Similarly, images taken at around 30% and 70% were also occluded by the contralateral leg for Plane 2 but are not shown above
This feature of biplane X-ray systems limits the types of joints and activities that can be imaged. For example, measurement of knee kinematics during walking has been limited to a portion of the gait cycle such as the stance phase, and most studies of lower-limb joint kinematics during locomotion have been confined to treadmill gait (Anderst et al. 2009).
Mobility of the X-Ray Imaging Systems Mobile biplane X-ray systems offer concurrent tracking and imaging of dynamic joint motion during human activity, which overcomes the limitations imposed by the small capture volume inherent to stationary systems. A mobile biplane X-ray (MoBiX) imaging system developed at the University of Melbourne translates each of the two X-ray units in a vertical plane so that the target joint remains within the image capture volume during movement (Fig. 4). The ability to track the motion of a target joint increases the size of the effective image capture volume of the system and enables the kinematics of various joints such as the knee, hip, and shoulder to be studied for a wide range of activities, including overground and treadmill walking, lunging, squatting, jumping, and stair ambulation. Section “Contrasting Mobile and Stationary Biplane Videoradiography Systems” below presents a more comprehensive discussion of the advantages and disadvantages of mobile and stationary biplane videoradiography systems.
Data Collection Key factors that must be considered during data collection include X-ray generator settings, specifically, tube current and voltage, and camera settings such as frame rate and exposure time. These settings are important not only because they collectively influence the quality of the images captured but also because they affect the radiation dosage received by the subject.
106
H. Gray et al.
Fig. 4 A mobile biplane X-ray (MoBiX) imaging system developed at the University of Melbourne comprising of two X-ray units mounted on a custom robotic gantry mechanism. Each X-ray unit comprising an X-ray tube and an image intensifier is mounted on the robotic arms that translate vertically along the mobile columns. The mobile columns translate along the horizontal guides positioned on either side of the walkway. The X-ray units translate both horizontally and vertically, thus increasing the effective image capture volume
X-Ray Generator Settings Appropriate values of X-ray tube current and voltage must be selected to produce images of sufficient brightness and contrast. In addition, the X-ray beam may be operated in either continuous or pulse mode. These parameters determine the subject’s exposure to ionizing radiation and are therefore carefully scrutinized by ethics committees to ensure the benefits outweigh the risks. A wide range of X-ray settings have been used for studies reported in the literature. In studies involving the knee joint, for example, Giphart et al. (2012) used a continuous voltage and current of 60 kV and 60 mA, respectively, while Anderst et al. (2009) used a voltage and current of 90 kV and 100 mA, respectively.
Frame Rate The upper limit in the frame rate of commercially available videoradiography systems is typically 30 Hz. However, in order to capture detailed 3D joint kinematics, many biplane videoradiography systems have been fitted with high-speed cameras capable of imaging at much higher frame rates, for example, the systems described by Guan et al. (2016) and Ivester et al. (2015) are capable of maximum frame rates of 1000 Hz. In general, faster activities require higher frame rates to
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
107
capture more detailed kinematic measurements. Frame rates used in recent studies involving biplane videoradiography systems have typically ranged from 30 Hz to 500 Hz (e.g., Li et al. 2009; Myers et al. 2012; Tashman et al. 2004).
Exposure Time Exposure time or shutter speed (measured in s) is the duration the camera sensor is exposed to light for a single image frame. Shorter exposure times produce sharper images as they reduce motion blur. However, longer exposure times allow more light to reach the sensor and therefore make the images brighter and have potential to improve image contrast. Exposure times in recent studies involving biplane videoradiography systems have ranged from 0.5 ms to 8 ms (e.g., Li et al. 2009; Myers et al. 2012).
Motion Blur Motion blur is caused by relative motion between the imaging system and the object and is proportional to the distance the object moves relative to the imaging system in the imaging plane during the exposure time. Therefore, motion blur is proportional to both exposure time and the relative velocity between the object and the imaging system. Two approaches have been used to reduce motion blur. Mobile biplane videoradiography systems reduce motion blur by reducing the relative velocity between the object and the imaging system. Other systems reduce motion blur by reducing the exposure time. Reducing exposure time reduces the light falling on the image sensor causing the images to be dark. This problem can be overcome by employing a higher X-ray current to maintain sufficient image brightness and contrast leading to an increase in the X-ray dosage received by the subject. However, some systems are able to pulse the X-ray beam synchronously with camera exposure time ensuring that the X-ray beam is only emitted by the X-ray generator when the camera shutter is open. This enables sufficiently bright images with little blur to be obtained while only exposing the subject to radiation when the joint is being imaged (Ivester et al. 2015; Tashman 2016).
Data Processing Image Preparation Raw images obtained from X-ray equipment are processed prior to pose estimation by porting the images through a pipeline comprised of flat field correction, distortion correction, and other image enhancement processors (Fig. 5). Flat field correction is performed to compensate for the differences in sensitivity between pixels in the
108
H. Gray et al.
Fig. 5 An example of an image processing pipeline comprising flat field correction, distortion correction, and feature enhancement
camera sensor. X-ray image distortion occurs within the image intensifier and the lens of the camera and may affect the measurement accuracy of the system. An image of an object embedded with an array of equispaced radiopaque beads is used to model the distortion as polynomial functions. These functions are then used to correct the distortion of the X-ray images (Garling et al. 2005). Further image processing is often performed to reduce noise and blur and to enhance features such as edges on the X-ray images.
Calibration Calibration of a biplane videoradiography system involves determining the geometric configuration of the X-ray sources and the image intensifiers. This procedure is accomplished by using the biplane system to image a calibration object containing several radiopaque beads at precisely known locations. The positions of the beads on the x-ray images are then used to calculate the geometric configurations of the imaging units (Kaptein et al. 2011).
Pose Estimation Pose estimation is the process of calculating the three translational and three rotational parameters needed to fully describe the position and orientation of each bone (or implant) in a common reference frame. This procedure is usually accomplished by solving an optimization problem that minimizes a scalar cost function. The unknown variables in this optimization problem are the six independent variables that define the pose of a geometric model of the bone or implant. The cost is formulated as a function of the six independent variables and reflects the accuracy with which the computed projections of the model can be superimposed onto the biplane X-ray images recorded in each image plane. An initial guess for the six variables is provided, and the cost function is minimized to calculate the pose. Various possibilities exist for the formulation of the cost function, for example, an edge-based method where root mean squared (RMS) distances between the detected edges on the
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
109
Fig. 6 DRR generation using ray tracing. For each view, a ray is traced from the X-ray source (green and red spheres) to every pixel in the simulated X-ray image. The value assigned to the pixel is the weighted sum of the voxel intensities of the bone models through which the ray passes. Each voxel is weighted by the distance travelled through it by the ray
X-ray images and the projected edges of the geometric model have been used by some researchers (Bingham and Li 2006; Guan et al. 2016), while others have used the correlation between X-ray images and digitally reconstructed radiographs (Bey et al. 2006; Ohnishi et al. 2010). Digitally reconstructed radiographs (DRRs) are generated by simulating the X-ray imaging process. A ray is traced from the X-ray source to each pixel on the imaging plane. The sum of intensities of the voxels in the bone model that a ray passes through is assigned to the respective pixel in the DRR (Fig. 6). To create the geometric model of an implant, a CAD model can usually be obtained from the manufacturer. For bones, segmented image data and 3D surface models are typically reconstructed from CT images. There are many commercial and open-source software packages available for segmenting individual bones from CT images and generating smooth polyhedral surface models (e.g., Mimics (Materialise N.V., Leuven, Belgium), 3D Slicer (Fedorov et al. 2016)). The segmented CT voxel data are used to create the DRRs. Pose Optimization: There are two general methods of optimizing the poses of bones or implants in biplane X-ray images. Descent-direction optimizers, such as the Levenberg-Marquardt algorithm, use the gradient of the cost function at the current pose to iteratively calculate a subsequent pose with a smaller value of cost function. The challenge with these methods is that they find the local minimum relative to the initial pose, which is not necessarily the optimal solution. Global optimizers, such as simulated annealing, use deterministic or stochastic methods to search the entire solution space for the global minimum of the cost function. The challenge with these algorithms is that they are computationally expensive and the value of the cost function at the global minimum is not known in advance. Thus it is not known when the algorithm should terminate, so choosing the number of iterations involves a
110
H. Gray et al.
balance between finding the optimal pose and finishing in a reasonable amount of time. Both of these challenges can be mitigated by providing an initial pose that is close to the optimal solution. Manually positioning the bone using a graphical user interface for each time frame can be a time-consuming and tedious process. A more efficient technique is to use the optimal pose from one or more previously solved frames to calculate an initial pose for the next frame. This procedure usually requires manual pose estimation to be performed only for the first frame or two. Another solution to this problem is to use traditional motion capture (surface markers, markerless, or inertial measurement units) to measure the approximate bone poses during the activity and to use these as the initial poses during optimization (Bone tracking software, C-Motion, MD, USA). 4D Tracking: An alternative to frame-by-frame pose estimation as described above is to optimize the pose for several time frames simultaneously, a method which we refer to here as 4D tracking (Tracking software, C-Motion, MD, USA). This process involves fitting splines to the six degrees of freedom of the initial bone poses calculated from the traditional motion capture data, with nodes at specified intervals (e.g., every fifth time frame), and treating the nodes as the independent variables during optimization. In each iteration, the optimization algorithm generates a set of node values and calculates the corresponding value of the cost function. Specifically, the algorithm generates DRRs for every X-ray image in the motion sequence for both views. The pose of the bone for each DRR is calculated by evaluating the node splines at the time of the X-ray image corresponding to that DRR. Each DRR is then compared to its X-ray image, and the correlation metrics for all DRR/X-ray pairs are summed to generate the value of the cost function for that set of node values. Natural cubic splines are used to interpolate the nodes, with a userdefined low-pass cutoff frequency for smoothing. 4D tracking has two primary advantages over a sequential process in which each time frame is solved independently. First, 4D tracking exploits temporal coherence in the data, which guarantees a smooth transition between successive poses and enforces reasonable motion physics. The smoothness of the splines and the spacing of the nodes can be controlled by the user and are usually matched to the activity being measured. Typical low-pass cutoff frequencies are 10 Hz for walking and 20 Hz for running. The second advantage of 4D tracking relates to the timing of the X-ray devices. Most current biplane videoradiography systems are capable of synchronous recording of the two X-ray views. With synchronous recording, the X-ray emitters are activated at the same time, and the two images are recorded simultaneously. This synchronicity eases the burden of pose estimation because it enables independent optimization for each time frame reducing the number of independent variables associated with 4D tracking. However, synchronous activation of the X-ray devices introduces additional noise into the system. When one X-ray beam intersects soft tissue or bone, some of its X-rays scatter and hit the image intensifier of the other view. When imaging parts of the body with small bones and a large amount of soft tissue, such as the lumbar spine, X-ray scatter can be a significant source of noise. With 4D tracking, the node splines can be evaluated at any time between the first and last frames (Fig. 7), and the optimization process can proceed whether the X-ray images are synchronous or asynchronous.
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
111
60.0
50.0
angle (deg.)
40.0
30.0
20.0
10.0
0.0 0.28
0.29
0.30
0.31 time (sec.)
0.32
0.33
Fig. 7 Example illustrating spline fitting one of the angles expressing the poses of an object for 4D tracking with asynchronous X-ray data. The green (X-ray Plane 1) and red (X-ray Plane 2) lines represent the frame times for the X-ray views. The circles represent the control points (nodes) of a spline, which is smoothed using a user-specified cutoff frequency. The spline is interpolated at each time frame (green and red diamonds) to evaluate the bone poses while tracking
Validation of Measurement Accuracy The errors involved in measuring 3D joint motion are dependent on the system as well as the target joint, activity, and the speed at which the task is performed. In order to evaluate measurement errors, the kinematic measurements obtained from the system need to be compared against a “gold standard” method. The most widely accepted procedure for obtaining accurate kinematic measurements involves embedding X-ray opaque beads in the bones and using their positions in the X-ray images to quantify joint kinematics. The ideal method for determining measurement accuracy of a biplane videoradiography system is described below (Fig. 8): 1. Insert at least three radiopaque beads in each of the bones of the target joint of a living person. The beads should be positioned nonlinearly and near enough to the target joint so that they are clearly visible in the X-ray images. 2. Collect biplane videoradiography images of the joint while the subject performs the activity of interest. 3. Calculate the joint kinematics using the locations of the beads on the X-ray images as described by You et al. (2001). 4. Calculate joint kinematics from biplane X-ray images using a markerless pose estimation method. 5. Compare the kinematics obtained from the markerless pose estimation method in step 4 with those obtained from the more accurate marker-based method in step 3.
112
H. Gray et al.
Fig. 8 Procedure involved in quantifying the measurement accuracy of a biplane videoradiography system (See text for details)
The aforementioned procedure for estimating measurement errors is invasive and thus rarely used. Several alternative variations to this method have been reported that involve inserting radiopaque beads into human cadaver joints, animal cadaver joints, live animal joints, and inanimate objects. Studies using human or animal cadaver joints are either conducted under static conditions or at significantly lower speeds than the activity intended to be studied (Ohnishi et al. 2010). Studies using live animals provide more realistic joint kinematics and speeds although differences in the 3D geometric shapes and sizes between the animal bones and human bones may make the results less convincing. Dynamic joint motion simulators (DJMS) overcome the above limitations as they enable validation studies to be performed on human cadaver joints under dynamic conditions which closely replicate the conditions present in vivo. For example, Guan et al. (2016) used a DJMS with an intact human cadaver knee to simulate joint motion during normal walking over ground.
Contrasting Mobile and Stationary Biplane Videoradiography Systems Perhaps most significantly, mobile biplane videoradiography systems offer a much larger effective image capture volume compared to stationary systems, enabling the study of a wider range of joints and activities. Another advantage of simultaneously tracking and imaging a joint is the potential reduction in radiation dosage to the subject. Radiation dosage, motion blur, and camera exposure time are interrelated as described earlier in this chapter. Because the MoBiX tracks the joint while imaging it, the relative motion between the joint and the imaging system is reduced, enabling
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
113
longer exposure times while keeping motion blur to an acceptable level. This feature enables X-ray currents to be reduced while still maintaining sufficiently bright images. However, mobile biplane systems are not without their limitations. They are more difficult and more costly to design and build than their stationary counterpart. Furthermore, the moving parts in a mobile system are subject to wear and tear and require regular maintenance. Misalignment of the robotic actuator guideways during installation and vibrations during operation both have the potential to change the relative positions of the X-ray tubes and image intensifiers away from their configuration at the time of system calibration. This problem can be minimized through precision manufacture and installation, using components with high rigidity and minimizing vibrations through improved joint tracking algorithms (Guan et al. 2016).
Conclusion and Future Directions Biplane videoradiography is currently the most accurate method for noninvasive measurement of dynamic in vivo joint kinematics during human movement. This chapter summarizes the basic concepts and considerations in configuring a biplane fluoroscopy system and accurately capturing and processing the X-ray image outputs. It also discusses some aspects related to validation of system accuracy and compares and contrasts stationary and mobile systems. The authors believe that the near future will see significant developments in hardware and software in a more integrated and streamlined manner making such systems and data processing more user-friendly and thus more accessible to clinicians and healthcare specialists. The accuracy of such systems may also continue to improve with the development of improved imaging systems with higher image quality in contrast and resolution. The development of more advanced CT scanners and videoradiography units capable of using lower radiation dosage may make these systems safer and thus suitable for studying a wider range of subject populations, including young children. It is hoped that these devices may be used ultimately as diagnostic tools for designing patientspecific treatment regimens for patients with various musculoskeletal conditions.
References Anderst W, Zauel R, Bishop J, Demps E, Tashman S (2009) Validation of three-dimensional modelbased tibio-femoral tracking during running. Med Eng Phys 31:10–16. https://doi.org/10.1016/j. medengphy.2008.03.003 Anderst WJ, Vaidya R, Tashman S (2008) A technique to measure three-dimensional in vivo rotation of fused and adjacent lumbar vertebrae. Spine J 8:991–997. https://doi.org/10.1016/j. spinee.2007.07.390 Banks SA, Hodge WA (1996) Accurate measurement of three dimensional knee replacement kinematics using single-plane fluoroscopy. IEEE Trans Biomed Eng 46:638–649 Bey MJ, Zauel R, Brock SK, Tashman S (2006) Validation of a new model-based tracking technique for measuring three-dimensional, in vivo glenohumeral joint kinematics. J Biomech Eng 128:604–609. https://doi.org/10.1115/1.2206199
114
H. Gray et al.
Bingham J, Li G (2006) An optimized image matching method for determining in-vivo TKA kinematics with a dual-orthogonal fluoroscopic imaging system. J Biomech Eng 128:588–595. https://doi.org/10.1115/1.2205865 Brainerd EL, Baier DB, Gatesy SM, Hedrick TL, Metzger KA, Gilbert SL, Crisco JJ (2010) X-ray reconstruction of moving morphology (XROMM): precision, accuracy and applications in comparative biomechanics research. J Exp Zool A Ecol Genet Physiol 313:262–279. https://doi.org/ 10.1002/jez.589 Davidson JM (1898) Roentgen rays and localisation: an apparatus for exact measurement and localisation by means of roentgen rays. Br Med J 1:10–13 Dennis DA, Mahfouz MR, Komistek RD, Hoff W (2005) In vivo determination of normal and anterior cruciate ligament-deficient knee kinematics. J Biomech 38:241–253. https://doi.org/ 10.1016/j.jbiomech.2004.02.042 Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin J-C, Pujol S, Bauer C, Jennings D, Fennessy F, Sonka M, Buatti J, Aylward S, Miller JV, Pieper S, Kikinis R (2016) 3D slicer as an image computing platform for the quantitative imaging network. Magn Reson Imaging 30:1323–1341. https://doi.org/10.1016/j.mri.2012.05.001 Garling EH, Kaptein BL, Geleijns K, Nelissen RGHH, Valstar ER (2005) Marker configuration model-based roentgen fluoroscopic analysis. J Biomech 38:893–901. https://doi.org/10.1016/j. jbiomech.2004.04.026 Giphart JE, Zirker CA, Myers CA, Pennington WW, LaPrade RF (2012) Accuracy of a contourbased biplane fluoroscopy technique for tracking knee joint kinematics of different speeds. J Biomech 45:2935–2938. https://doi.org/10.1016/j.jbiomech.2012.08.045 Grood ES, Suntay WJ (1983) A joint coordinate system for the clinical description of threedimensional motions: application to the knee. J Biomech Eng 105:136–144 Guan S, Gray HA, Keynejad F, Pandy MG (2016) Mobile biplane X-ray imaging system for measuring 3D dynamic joint motion during overground gait. Med. Imaging IEEE Trans 35:326–336. https://doi.org/10.1109/TMI.2015.2473168 Ivester JC, Cyr AJ, Harris MD, Kulis MJ, Rullkoetter PJ, Shelburne KB (2015) A reconfigurable high-speed stereo-radiography system for sub-millimeter measurement of in vivo joint kinematics. J Med Device 9:41009. https://doi.org/10.1115/1.4030778 Kaptein BL, Shelburne KB, Torry MR, Giphart JE (2011) A comparison of calibration methods for stereo fluoroscopic imaging systems. J Biomech 44:2511–2515. https://doi.org/10.1016/j. jbiomech.2011.07.001 Kärrholm J, Selvik G, Elmqvist L-G, Hansson LI (1988) Active knee motion after cruciate ligament rupture. Acta Orthop Scand 59:158–164. https://doi.org/10.1080/17453678809169699 Li G, Kozanek M, Hosseini A, Liu F, Van de Velde SK, Rubash HE (2009) New fluoroscopic imaging technique for investigation of 6DOF knee kinematics during treadmill gait. J Orthop Surg Res 4:6. https://doi.org/10.1186/1749-799X-4-6 List R, Foresti M, Gerber H, Goldhahn J, Rippstein P, Stussi E (2012) Three-dimensional kinematics of an unconstrained ankle arthroplasty: a preliminary in vivo videofluoroscopic feasibility study. Foot Ankle Int 33:883–892. https://doi.org/10.3113/FAI.2012.0883 Matteri RE, Pope MH, Frymoyer JW (1976) A biplane radiographic method of determining vertebral rotation in postmortem specimens. Clin Orthop Relat Res 116 Myers C a, Torry MR, Shelburne KB, Giphart JE, LaPrade RF, Woo SL-Y, Steadman JR (2012) In vivo tibiofemoral kinematics during four functional tasks of increasing demand using biplane fluoroscopy. Am J Sports Med 40:170–178. https://doi.org/10.1177/03635465 11423746 Ohnishi T, Suzuki M, Nawata A, Naomoto S, Iwasaki T, Haneishi H (2010) Three-dimensional motion study of femur, tibia, and patella at the knee joint from bi-plane fluoroscopy and CT images. Radiol Phys Technol 3:151–158. https://doi.org/10.1007/s12194-010-0090-1 Selvik G (1974) Roentgen stereophotogrammetry: a method for the study of the kinematics of the skeletal system. University of Lund, Lund. https://doi.org/10.3109/17453678909154184
Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography
115
Tashman S (2016) Comments on “validation of a non-invasive fluoroscopic imaging technique for the measurement of dynamic knee joint motion”. J Biomech 41:3290–3291. https://doi.org/ 10.1016/j.jbiomech.2008.07.038 Tashman S, Collon D, Anderson K, Kolowich P, Anderst W (2004) Abnormal rotational knee motion during running after anterior cruciate ligament reconstruction. Am J Sports Med 32:975–983. https://doi.org/10.1177/0363546503261709 Tashman S, Kolowich P, Collon D, Anderson K, Anderst W (2007) Dynamic function of the ACL-reconstructed knee during running. Clin Orthop Relat Res 454:66–73. https://doi.org/ 10.1097/BLO.0b013e31802bab3e Uvehammer J, Karrholm J, Brandsson S, Herberts P et al (2000) In vivo kinematics of total knee arthroplasty: flat compared with concave tibial joint surface. J Orthop Res 18:856–864 van Dijk R, Huiskes R, Selvik G (1979) Roentgen stereophotogrammetric methods for the evaluation of the three dimensional kinematic behaviour and cruciate ligament length patterns of the human knee joint. J Biomech 12:727–731. https://doi.org/10.1016/0021-9290(79)90021-6 You BM, Siy P, Anderst W, Tashman S (2001) In vivo measurement of 3D skeletal kinematics from sequences of biplane radiographs: application to knee kinematics. IEEE Trans Med Imaging 20:514–525. https://doi.org/10.1109/42.929617 Zhu Z, Li G (2011) An automatic 2D–3D image matching method for reproducing spatial knee joint positions using single or dual fluoroscopic images. Comput Methods Biomech Biomed Engin:1–12. https://doi.org/10.1080/10255842.2011.597387
3D Musculoskeletal Kinematics Using Dynamic MRI Frances T. Sheehan and Richard M. Smith
Abstract
Until the early 1990s, the tools available to measure musculoskeletal motion were typically highly invasive. Thus, knowledge of musculoskeletal system dynamics was primarily derived through cadaver and modeling experiments. The rapid development of dynamic magnetic resonance (MR) imaging techniques changed this and opened vast new opportunities for the study of 3D musculoskeletal dynamics during volitional activities. Today, dynamic MR methodologies remain the only techniques that can noninvasively track in vivo 3D musculoskeletal movement. One difficulty in applying these dynamic MR techniques to the study of musculoskeletal motion is the complex interplay of parameters that affect the spatial/temporal resolution, accuracy, and precision. The purpose of this chapter is to first provide an explanation of the fundamental principles behind two of these dynamic imaging techniques, cine and cine phase-contrast MR. Tagged cine MR is another technique that has been primarily used to track muscle motion and strain but will not be addressed. In doing so, this will create a platform for future experimental designs using dynamic MR. This will be followed by a review of the accuracies, the advantages, and disadvantages of the these dynamic MR methods. Finally, several previously published studies will be highlighted to provide an explanation of how these techniques can be applied and what main challenges must be considered for future experiments using dynamic MR.
F.T. Sheehan (*) • R.M. Smith Rehabilitation Medicine Department, Functional and Applied Biomechanics Section, National Institutes of Health, Bethesda, MD, USA e-mail: [email protected]; [email protected]; [email protected] # This is a U.S. Government work and not under copyright protection in the US; foreign copyright protection may apply 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_155
117
118
F.T. Sheehan and R.M. Smith
Keywords
MR • Magnetic resonance imaging • Cine • Cine phase contrast • Fastcard • CPC • Fast-PC • Musculoskeletal • Kinematics • Strain • Moment arms • Tendon
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Static MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cine MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fastcard (Fast Cine) MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fast Cine: Phase-Contrast (Fast-PC) MR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advanced Applications of Dynamic MR Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combining Fastcard Imaging with 3D Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applications of CPC and Fast-PC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
118 119 119 121 121 122 123 126 126 128 128 129 130
Introduction Until the early 1990s, the tools available to measure musculoskeletal motion were typically highly invasive (Regev et al. 2011; Edsfeldt et al. 2015; Lafortune et al. 1994; van Kampen and Huiskes 1990; Manal et al. 2000). Thus, knowledge of musculoskeletal system dynamics was primarily derived through cadaver and modeling experiments. The rapid development of dynamic magnetic resonance (MR) imaging techniques changed this and opened vast new opportunities for the study of 3D musculoskeletal dynamics during volitional activities (Burnett et al. 1987; Drace and Pelc 1994; Sheehan et al. 1998; Sheehan and Drace 2000; Asakawa et al. 2002; Pappas et al. 2002; Fujiwara et al. 2004; Finni et al. 2006; Cheng et al. 2008). Today, dynamic MR methodologies remain the only techniques that can noninvasively track in vivo 3D musculoskeletal movement. One difficulty in applying these dynamic MR techniques to the study of musculoskeletal motion is the complex interplay of parameters that affect the spatial/ temporal resolution, accuracy, and precision. The purpose of this chapter is to first provide an explanation of the fundamental principles behind two of these dynamic imaging techniques, cine (Glover and Pelc 1988) and cine phase-contrast (Feinberg et al. 1984; Wedeen et al. 1985; Pelc et al. 1991b, 1994) MR. Tagged cine MR is another technique that has been primarily used to track muscle motion and strain but will not be addressed (Moerman et al. 2012). In doing so, this will create a platform for future experimental designs using dynamic MR. This will be followed by a review of the accuracies, the advantages, and disadvantages of the these dynamic MR methods. Finally, several previously published studies will be highlighted to provide an explanation of how these techniques can be applied and what main challenges must be considered for future experiments using dynamic MR.
3D Musculoskeletal Kinematics Using Dynamic MRI
119
State of the Art The earliest use of MR technology was isolated to spectroscopy (NRM or nuclear magnetic resonance). Once researchers realized how to manipulate the MR signal to produce in vivo images of the human tissue, the technology was rapidly transferred to the clinical domain in the early 1980s, providing a wealth of new information. A primary drawback of this new technology was the acquisition time for a single image prohibited its use in imaging moving tissue, particularly the heart. Cine MR was developed to acquire anatomic images of moving tissue during a cyclic movement sequence by synchronizing the data collection to the cardiac cycle. This was rapidly followed by the development of tagged MR and cine phase-contrast (CPC) MR, which allowed the analytical tracking of tissue. Tagged MR has remained a tool primarily used for cardiac imaging, but both cine and CPC MR have been widely applied to the noninvasive study of 3D musculoskeletal kinematics. The original applications of cine and CPC MR to the study of musculoskeletal motions required long acquisition times with a high number of repeated cycles. Currently, the accuracy of using CPC to track musculoskeletal motion has dropped to less than 0.3 mm. In addition, the imaging times have dramatically dropped, enabling acquisitions of less than a minute.
Static MR Imaging MR imaging is distinct from other imaging techniques in two key ways. First, the measured signal is generated from the tissue being imaged. MR imaging is based on the principle that atoms with an odd number of neutrons and protons will precess about an external magnetic field at a frequency directly proportional to the field strength. Clinical MR imaging typically focusses on hydrogen atoms, often referred to as “spins.” With the introduction of a perpendicular magnetic field at the same frequency of precession, the spins will “tip” into the transverse plane, creating a transverse magnetization. The MR signal is the transverse magnetization of all spins within a voxel (the smallest volumetric unit of the image). As the signal for all voxels is collected together in a single summed signal, the location of a voxel’s signal is encoded in the phase and frequency of its signal using magnetic gradients (Riederer 1993). This encoding process results in another unique feature of MR imaging; the data collected during a scan is not spatial but spatial frequency data (k-space data, Fig. 1). More importantly, numerous data acquisitions are required to provide a complete spatial frequency map (k-space map) that can be converted into an image using inverse Fourier transforms. Various methods for collecting data in k-space have been developed (Foo et al. 1995; Markl et al. 2003; Pike et al. 1994; Thompson and McVeigh 2004; Thunberg et al. 2003), but the simplest, collecting a single line in k-space (phase encode) per acquisition (Fig. 2), will be used as the example for this chapter.
120
F.T. Sheehan and R.M. Smith
Fig. 1 K-space map to MR image. K-space map (left) of a 2D GRE sagittal image of the knee (right). The majority of the data in the k-space map resides at the lower frequencies, which represent the portions of the image where the contrast is not changing rapidly from one pixel to the next. For example, the magnitude at kx = ky = 0 represent the parts of the 2D spatial image where the contrast is not changing from pixel to pixel. One such region of low spatial frequency is shown on the left by the number 1. The number 2 denotes a region of high spatial frequency in the anteriorposterior direction
Fig. 2 K-space map representing a linear phase encode collection: in this example, the frequency direction is kx (anteriorposterior) and the phase direction is ky (superiorinferior). For visual clarity, the acquisition is represented with very low frequency and phase resolution (41 41). For this example, 41 individual phase encodes (each represented by a unique color) are required to produce one 2D spatial image
3D Musculoskeletal Kinematics Using Dynamic MRI
121
The speed at which a single phase encode can be captured is referred to as the repetition rate (TR). Thus, the time needed to acquire an entire image is: Imaging time ¼ #phases TR
(1)
#phases = number of required phase encodes to create a single image If data averaging is used, to reduce the effect of random noise, then the imaging times increase linearly with the number of averages (NA) acquired: Imaging time ¼ NA #phases TR
(2)
As noted above (Eqs. 1 and 2), the imaging time is dependent on the number of phase encodes acquired. For a fixed field of view, #phases directly determines the spatial resolution of the final image. To put numbers to this, let us assume a 256 256 (frequency resolution #phases) acquisition. This would result in a square image 256 pixels wide by 256 high. Assuming a TR = 5 ms and no data averaging, the imaging time (Eq. 2) would be 1.28 s for a single 2D image. If we set the spatial field of view at 200 200 mm, the spatial resolution becomes (200 mm/ 256) = 0.78 mm/pixel.
Dynamic MR Imaging Cine MR When the first scanners became commercially available in the early 1980s, it was not feasible to image the beating heart, because its period of motion was significantly greater than the image acquisition time. As noted in the example above for a 256 256 acquisition with a TR of 5 ms, the imaging time (1.28 s) is longer than the typical cardiac period of 1 s. A TR of 5 ms has only been feasible recently, when cine MR was first being developed TRs above 20 ms were typical (Keegan et al. 1994; Sheehan et al. 1998). Cine MR (Waterton et al. 1985; Burnett et al. 1987; Glover and Pelc 1988; Cadera et al. 1992) was developed to overcome this limitation, with a particular focus on cardiac imaging. The underlying assumption of cine MR is that the tissue being imaged is moving in a repeatable, cyclic motion path. By gating the data collection to the cardiac cycle, the required phase encodes in k-space (256 in our example) can be acquired over numerous cardiac cycles. Retrospective gating with cine MR compensates for variations in the period of motion by repeatedly acquiring the same k-space data line every TR during a single cycle. A trigger is used to detect when a new cycle begins, signaling a change in the k-space data line being acquired (Fig. 3). After #phases cycles, the data are retrospectively interpolated so that images are created at specified time intervals. Although any number of image time frames can be created through the interpolation process, the true temporal resolution is TR, whereas, the spatial resolution is dependent on #phases:
122
F.T. Sheehan and R.M. Smith
Fig. 3 Cine MR data capture: when the scanner receives a trigger that a new cycle has begun (black line on red signal), the ky data line (phase encode) being acquired is incremented. In the above example, all cycles are of even length and a true 16 frames is captured. The temporal resolution is equal to the TR (time between data captures). Compensating for variations in the periods of each cycle is accomplished with retrospective gating and interpolation to align the data to the correct temporal location in the motion cycle
cine temporal resolution ¼ TR
(3)
spatial resolution ¼ FOV=#phases
(4)
Thus, with a TR of 5 ms, 200 true frames of data are acquired with a period of 1 s. Other types of gating, including prospective, are available for cine MR. One issue with prospective gating is that the entire cycle is typically not captured. The imaging time for cine MR depends on the #phases and motion period (T) of the moving tissue being imaged: cine imaging time ¼ T #phases
(5)
Following the example above, and assuming T = 1 s, the cine imaging time = 1 s*256 = 4.3 min (Eq. 5), requiring 256 cycles of motion. Although the cine techniques were motivated by a desire to image cardiac motion, its applicability to musculoskeletal motion was quickly realized. In 1987, a study using cine MR to visualize passive temporomandibular joint mechanics (Burnett et al. 1987) was published. This was quickly followed by a qualitative study of active ankle joint motion (Melchert et al. 1992) and a quantitative study of active knee joint motion (Brossmann et al. 1993). More recently, cine MR has been used to evaluate tongue movement (Stone et al. 2001), fetal motion (Verbruggen et al. 2016), and scaphoid-lunate motion (Langner et al. 2015).
Fastcard (Fast Cine) MR Fastcard (Foo et al. 1995), or fast cine MR, was developed to enable a trade-off between temporal resolution and the number of cycles required to capture a full image set. Instead of repeatedly capturing a single phase encode line per motion
3D Musculoskeletal Kinematics Using Dynamic MRI
123
cycle, fastcard collects multiple phase encodes during a single motion cycle. This reduces the overall imaging time, with a degradation in temporal resolution: fastcard imaging time ¼ T #phases=#views
(6)
fastcard temporal resolution ¼ #views TR
(7)
#views = the number of unique phase encode lines acquired during a single motion cycle For most joint motion studies using cine MR, the period of motion is longer than that of a beating heart. Using the above example, but assuming a typical knee joint extension/flexion type movement (Brossmann et al. 1993), with a motion period of 2 s, the 256 motion cycles would require 8.5 min to capture a cine MR image set. Yet, 400 temporal frames (a temporal resolution of 5 ms) would be produced to represent an arc of motion of approximately 40 . Using fastcard and assuming 20 views are collected during each motion cycle, the fastcard imaging time is just 26 s, requiring only 13 motion cycles. Here, only 20 true data frames are collected, with a temporal resolution equal to 100 ms. This trade-off between temporal resolution and number of required motion cycles afforded by fastcard is fundamental to the design of experiments that can be repeatedly performed by both healthy volunteers and individuals with musculoskeletal impairments/pathologies. Fastcard also enables a trade-off between motion period and the number of required motion cycles. Using the same example, if the period is increased by a factor of 2 (T = 4), the number of views acquired during each motion cycle could be doubled (40 views). Although the temporal resolution would increase by a factor of 2, the speed of motion would reduce by the same amount. As such, no increase in temporal blurring would be expected. This would reduce the number of required motion cycles to just 7, but the overall acquisition time would still remain at ~26 s. It is important to note that this example is assuming an acquisition matrix of 256 256. Using the spatial parameters from a previous study (Carlson et al. 2016) of patellofemoral kinematics (FOV = 180 mm and using a 75% phase acquisition), the acquisition matrix can be dropped to 180 135 pixels and still maintain a resolution of 1 1 mm. This would require just four motion cycles with an imaging time of 16 s. Such a reduced number of required cycles could potentially allow for studies with increased joint loading and is crucial to acquiring data during volitional joint motion in individuals with musculoskeletal pathologies/impairments (Fig. 4).
Fast Cine: Phase-Contrast (Fast-PC) MR CPC MR (Pelc et al. 1991b) combines the ability of phase-contrast imaging to produce three-dimensional quantitative velocity data with cine’s ability to produce a series of images throughout a gated motion cycle. Fast-PC MR (Foo et al. 1995) is identical to CPC MR but inherits fastcard’s ability to capture multiple views, or phase encode lines, during a single motion/cardiac cycle. Phase-contrast imaging
124
F.T. Sheehan and R.M. Smith
Fig. 4 Fastcard data acquisition: for this figure, five phase encode lines (#views = 5) are captured per motion cycle. Thus, phase encode lines ky = 16 through ky = 20 are captured during the first motion cycle. Temporal resolution is increased to 5*TR (the time between acquiring the same k-space line is 5*TR). Yet, on the positive side, the imaging time and the number of cycles required are also reduced by a factor of 5
Fig. 5 Acquiring fast-PC data. To capture the full fast-PC dataset, four separate data acquisitions are needed. This provides measures of velocity in the frequency (x), phase (y), and slice (z) directions. The temporal resolution is reduced by a factor of 4 (temporal resolution = 4 * TR) and 5 true frames of data are available. If the #views was increased from 1 to 2, then the temporal resolution would increase to 8*TR, with a reduction by a factor of 2 in the number of motion cycles and the imaging time
(Pelc 1995; Pelc et al. 1991a) manipulates the concept that the MR signal from spins moving in the direction of a magnetic gradient will accumulate phase proportional to the first moment of that gradient. In MR imaging, magnetic gradient fields are used in three perpendicular directions to isolate the slice being imaged (slice selection gradient) and to encode signal’s location into the phase and frequency of the signal (phase and frequency gradients) (Riederer 1993). Thus, for PC imaging, each of these gradients is sequentially modified to enable data an acquisition that is sensitive to velocity in the three perpendicular directions, frequency, phase, and slice (Fig. 5). In addition, a fourth acquisition is used that is insensitive to motion, as a reference. Thus, for each phase encode line in k-space, four acquisitions are needed to capture a
3D Musculoskeletal Kinematics Using Dynamic MRI
125
Fig. 6 Fast-PC acquisition of the knee (Sheehan et al. 2012): from top to bottom, the rows represent the anatomic images and the right-left (RL), the anterior-posterior (AP), and the superior-inferior (SI) velocity images. In the velocity images, pure white represents +30 cm/s and black represents 30 cm/s. The following parameters were used: TR = 5 ms, motion rate = 30 cycles/min; phase direction = AP; #views = 4; number of averages = 2; percent phase fov = 72%; 256 256 pixels; 200 mm fov, acquired resolution = 0.94 0.94 8 mm; reconstructed resolution = 0.78 0.78 8 mm; maximum velocity encoding = 30 cm/s. The temporal resolution = 80 ms = 25 frames of data (Eq. 8). Imaging time = 1 min 30 s, 45 motion cycles (Eq. 9). Note, the scan time and number of required cycles could easily be reduced by 50% if average was not used, but all subjects could tolerate the scan time and the data averaging reduced random noise in the images. The black edges seen on the right and left side of the images are due to the 75% phase fov, which reduced the imaging time by 25%, without reducing the spatial resolution
complete (velocity in three perpendicular direction) CPC data line. This increases the overall temporal resolution by a factor of 4, with no reduction in imaging time or number of cycles required: FastPC temporal resolution ¼ 4 TR #views
(8)
FastPC imaging time ¼ T #phases=#views
(9)
The complete acquisition produces a temporal series of images representing the anatomy within the imaging plane, plus images representing the velocity in the frequency, phase, and slice (x, y, and z) directions (Figs. 5 and 6). Thus, following from above, using a FOV = 180 mm, a 75% phase acquisition, a T = 2 s, a TR = 5 ms,
126
F.T. Sheehan and R.M. Smith
and 20 views, the temporal resolution would be 400 ms, allowing for only 5 true frames of data. By reducing the #views to 5, the temporal resolution returns to 100 ms with 20 true frames of data. Yet, this comes at a cost in imaging time, which would increase to 54 s. A final key parameter when using fast-PC MR to track musculoskeletal motion is the maximum velocity encoding, venc. As the velocity is encoded in the phase of the signal, a phase shift of 180 is designated as the venc. Any velocity greater than 180 will create velocity aliasing. For example, a velocity producing a phase shift of 270 could not be distinguished from a negative velocity that produced a 90 phase shift. Thus, keeping venc low will improve the velocity resolution, but potentially could create aliasing. Further, any reduction in venc results in increases in TR, which will negatively affect the temporal resolution. For both cine and CPC imaging, any tissue that is moving in the imaging plane that is not synchronized to the data capture will result in destructive noise (ghosts). The largest sources of such noise when using dynamic MR to track motion are inconsistent movement and blood flow within the image. The noise from the latter can be greatly reduced using spatial pre-saturation of the blood (Im et al. 2015; Wood and Wiang 1993). This pre-saturation destroys the MR signal in the blood prior to it entering the imaging plane, thus eliminating (or greatly reducing) the noise it generates.
Advanced Applications of Dynamic MR Imaging Combining Fastcard Imaging with 3D Modeling One disadvantage of cine MR is that quantifying musculoskeletal movement from 2D spatial images becomes less accurate when there is out-of-plane motion, as the points of reference cannot be directly tracked throughout the entire movement, which could lead to larger errors (Shibanuma et al. 2004, 2005). Yet, with 2D fastcard imaging times approaching just 13 s and newer low-resolution cine imaging becoming available (Kaiser et al. 2013), it is quite possible to capture a multiplane fastcard (MPC) image set with limited scanning time. Scan times from 1.5 min (45 cycles) (Borotikar et al. 2012) to 5 min (150 cycles) (Kaiser et al. 2016) have been reported. From each time frame, a low-resolution model of the bone being tracked can be created and fit to a static 3D model of the bone, providing a method to consistently track the 3D motion of the bone (Fig. 7). This model fitting methodology also provides a method for quantifying dynamic cartilage contact parameters (Borotikar et al. 2012; Kaiser et al. 2016). Model fitting using dynamic MR data has been applied in two ways. The first is based on fitting the sparse dynamic model from a single time frame to a 3D static model of the same bone (Borotikar et al. 2012). From a single time frame in the MPC image set, a sparse model of the bone is created. Next, a high-resolution model of the same bone is created from static images. The optimal rotation and translation that minimize errors between the surfaces of the two models of the same bone are found.
3D Musculoskeletal Kinematics Using Dynamic MRI
127
Fig. 7 3D kinematics derived from combining fastcard imaging and 3D modeling. This example set is based on a previous study (Borotikar et al. 2012), where fastcard imaging was used to produce a 4D image set representing 24 evenly spaced temporal increments throughout the motion (first column). Each 3D image (representing the anatomy at a single time frame) is comprised of seven images. A sparse dynamic model of the bone (2nd column) is extracted from each of the 24, 3D image sets. The femur is shown segmented, but this can be done for the patella and tibia as well. From a high-resolution 3D image static image set, a high-resolution model of the bone is created (third column). Using registration, typically iterative closest point algorithm, the rotation and translation of the rough dynamic model that places the dynamic and static model into the best alignment are calculated. This is independently accomplished at each time frame. These rotation and translation are expressed as a matrix (STT1 = transformation required to bring the dynamic model from time frame 1 into alignment with the static model). When this is completed for all time frames, the 3D kinematic motion of the bone is known (T1TTi). Applying these transforms to rigid models of the bone that include the cartilage enables an analysis of cartilage contact (fourth column)
Thus, the alignment of the 3D static model is now known for one dynamic time frame. The 3D displacement and rotation of the bone, derived from the fast-PC data, are then applied to the static model, so that its position is known for all time frames. This was done for the patellofemoral joint with an accuracy of less than 0.9 mm (average absolute error) reported (Borotikar et al. 2012). In addition, it is possible to create the sparse model for every time frame of interest (Kaiser et al. 2016; Borotikar et al. 2012). By fitting each sparse dynamic bone model to the 3D static bone model, the 3D kinematics of the bone can be backed out from the transformation matrices describing the sparse dynamic to high-resolution static model. This has been done for the tibiofemoral joint, with an accuracy of than 0.60 mm (RMS error)
128
F.T. Sheehan and R.M. Smith
(Kaiser et al. 2016), and for the patellofemoral joint with an accuracy less than 1.3 mm (average absolute error) (Borotikar et al. 2012). Ultimately, the final accuracy is affected by the signal-to-noise ratio, the spatial resolution of the images, the size and shape of the bone being tracked, and the type of motion being analyzed.
Applications of CPC and Fast-PC The patellofemoral and tibiofemoral joint dynamics were the first to be studied using CPC MR. With the addition of fast-PC, and continuously decreasing TRs, the time for acquiring a full CPC dataset of the knee has dropped from over 7 min (Sheehan et al. 1998) to just 1.5 min (Behnam et al. 2011). If data averaging had not been used, both of these times would be halved. This reduction in time opens up the use of CPC, as the number of required motion cycles dropped from just over 200 to 45 (motion rate = 30 cycles/min). During this time, improvements in scanner strength and coil design greatly enhanced the signal-to-noise ratio, which allowed an improved accuracy. Based on a nearly identical phantom experiment, the average absolute error for tracking both in- and out-of-plane motion dropped from a maximum 1.48 mm (Sheehan et al. 1998) to 0.33 mm (Behnam et al. 2011). A recent study by Jensen and colleagues (Jensen et al. 2015) did an excellent job exploring the various potential sources of error in tracking muscle strain with CPC, finding that the bias (average error) in the velocity measures was below 1.3 mm/s. In terms of tracking muscle, researchers have focused primarily on two types of studies. One is to track a single point on the muscle in order to calculate moment arms (Finni et al. 2006; Im et al. 2015; Sheehan 2012; Westphal et al. 2013; Wilson and Sheehan 2009), tendon paths (Wilson and Sheehan 2010), and muscle excursions (Wen et al. 2008). The other is to evaluate muscle deformation, particularly as a marker of pathology (Finni et al. 2006; Kinugasa et al. 2008; Pappas et al. 2002; Silder et al. 2010; Sinha et al. 2012; Zhou and Novotny 2007). A recent group has shown the feasibility of evaluating three-dimensional strain (Jensen et al. 2016). Unfortunately, the scan time for the 3D CPC images required for 3D strain measures is unrealistically long to be applied to the evaluation of in vivo muscle strain during a volitional activity. Yet, as scanners further improve, the available TRs continue to drop, and new 3D fast MR data acquisition algorithms are developed; the scan time will likely reduce enough to allow in vivo experiments.
Future Directions CPC and fast-PC MR remain the only methodologies available to noninvasively measure in vivo 3D musculoskeletal kinematics (Bey et al. 2008; Shih et al. 2003; Fregly et al. 2005; Manal et al. 2000; Moro-oka et al. 2007; Yamashita et al. 2007). The reported accuracy of tracking rigid motion (0.33 mm) is better than other noninvasive techniques that measure in vivo skeletal kinematics (e.g., fluoroscopy
3D Musculoskeletal Kinematics Using Dynamic MRI
129
and motion capture) and muscle dynamics (e.g., ultrasound). It has an advantage over x-ray-based techniques (e.g., fluoroscopy and 4D ultrasound) in that it does not expose subjects to ionizing radiation. Quantifying kinematics by combining model fitting with fastcard data is less accurate (Kaiser et al. 2016; Borotikar et al. 2012) than fast-PC imaging but does provide the opportunity to quantify cartilage contact. The accuracy of tracking musculoskeletal motion from a fast-PC acquisition is dependent on the strength of the magnet, the consistency of movement, the quality of the scanner, the TR, the number of data averages, the fov, the venc, and the signal-tonoise ratio within the imaging plane. The integration algorithms used to track the motion can help compensate for some of the errors due to noise and systematic errors from the MR scanner (Jensen et al. 2015; Pelc et al. 1995; Zhu et al. 1996). Thus, relying on past validation studies does provide a general framework for the accuracy of tracking musculoskeletal motion, but a validation based on the specific parameters being used for a particular study should be done in order to insure that the combination of parameters used leads to accurate tracking. As with all measurement techniques, dynamic MR imaging has its limitations. First, it relies on a costly imaging modality that is not readily available to all clinicians and researchers. For the most accurate data, the majority of studies use high field strength, closed-bore, MR units, which limits the types of movements that can be studied. Open-bore units are available and some cine studies have used this technology, but currently the images produced are of an inferior quality. Thus, as open-bore technology improves, the types of functional movements that can be evaluated will expand. Lastly, the dynamic MR techniques rely on repetitive movements. As acquisition methods and scanners continue to improve, this limitation is quickly being removed, and real-time dynamic MR imaging is becoming a reality. Another limitation is that the integration routines to track musculoskeletal motion are not widely available for all researchers. Yet, this will likely rapidly change in the next few years. Future advancements in various aspects of MRI technology will help expand the application of dynamic MR in the quantification of 3D musculoskeletal dynamics. Improvements in scanner and coil design will support a wider range of tasks that can be studied. Eventually, this will likely include the full range of motion for joints such as the shoulder, as well as the ability to study the kinematics of multiple joints during dynamic tasks. The development and sharing of integration algorithms and packages for CPC will support its expanded use in both the research and clinical setting. As the accuracy of real-time MR imaging reaches the level of the current CPC techniques, the types of motions and pathologies that can be studied will rapidly expand.
Cross-References ▶ Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion
130
F.T. Sheehan and R.M. Smith
Acknowledgments This work was funded by the Intramural Research Program of the National Institutes of Health Clinical Center, Bethesda, MD, USA. This research was also made possible through the NIH Medical Research Scholars Program, a public-private partnership (http://fnih.org). We thank Judith Welsh for her help and support in the work.
References Asakawa DS, Blemker SS, Gold GE, Delp SL (2002) In vivo motion of the rectus femoris muscle after tendon transfer surgery. J Biomech 35(8):1029–1037 Behnam AJ, Herzka DA, Sheehan FT (2011) Assessing the accuracy and precision of musculoskeletal motion tracking using cine-PC MRI on a 3.0T platform. J Biomech 44(1):193–197. https://doi.org/10.1016/j.jbiomech.2010.08.029 Bey MJ, Kline SK, Tashman S, Zauel R (2008) Accuracy of biplane x-ray imaging combined with model-based tracking for measuring in-vivo patellofemoral joint motion. J Orthop Surg Res 3:38. https://doi.org/10.1186/1749-799x-3-38 Borotikar BS, Sipprell WH 3rd, Wible EE, Sheehan FT (2012) A methodology to accurately quantify patellofemoral cartilage contact kinematics by combining 3D image shape registration and cine-PC MRI velocity data. J Biomech 45(6):1117–1122 Brossmann J, Muhle C, Schroder C, Melchert UH, Bull CC, Spielmann RP, Heller M (1993) Patellar tracking patterns during active and passive knee extension: evaluation with motiontriggered cine MR imaging. Radiology 187(1):205–212. https://doi.org/10.1148/radiology.187. 1.8451415 Burnett KR, Davis CL, Read J (1987) Dynamic display of the temporomandibular joint meniscus by using “fast-scan” MR imaging. AJR Am J Roentgenol 149(5):959–962. https://doi.org/10.2214/ ajr.149.5.959 Cadera W, Viirre E, Karlik S (1992) Cine magnetic resonance imaging of ocular motility. J Pediatr Ophthalmol Strabismus 29(2):120–122 Carlson VR, Boden BP, Sheehan FT (2016) Patellofemoral kinematics and tibial tuberositytrochlear groove distances in female adolescents with patellofemoral pain. Am J Sports Med. https://doi.org/10.1177/0363546516679139 Cheng S, Butler JE, Gandevia SC, Bilston LE (2008) Movement of the tongue during normal breathing in awake healthy humans. J Physiol 586(17):4283–4294. https://doi.org/10.1113/ jphysiol.2008.156430 Drace JE, Pelc NJ (1994) Skeletal muscle contraction: analysis with use of velocity distributions from phase-contrast MR imaging. Radiology 193(2):423–429. https://doi.org/10.1148/ radiology.193.2.7972757 Edsfeldt S, Rempel D, Kursa K, Diao E, Lattanza L (2015) In vivo flexor tendon forces generated during different rehabilitation exercises. J Hand Surg Eur Vol 40(7):705–710. https://doi.org/ 10.1177/1753193415591491 Feinberg DA, Crooks L, Hoenninger J 3rd, Arakawa M, Watts J (1984) Pulsatile blood velocity in human arteries displayed by magnetic resonance imaging. Radiology 153(1):177–180. https:// doi.org/10.1148/radiology.153.1.6473779 Finni T, Hodgson JA, Lai AM, Edgerton VR, Sinha S (2006) Muscle synergism during isometric plantarflexion in achilles tendon rupture patients and in normal subjects revealed by velocityencoded cine phase-contrast MRI. Clin Biomech (Bristol, Avon) 21(1):67–74. https://doi.org/ 10.1016/j.clinbiomech.2005.08.007 Foo TK, Bernstein MA, Aisen AM, Hernandez RJ, Collick BD, Bernstein T (1995) Improved ejection fraction and flow velocity estimates with use of view sharing and uniform repetition time excitation with fast cardiac techniques. Radiology 195(2):471–478. https://doi.org/ 10.1148/radiology.195.2.7724769
3D Musculoskeletal Kinematics Using Dynamic MRI
131
Fregly BJ, Rahman HA, Banks SA (2005) Theoretical accuracy of model-based shape matching for measuring natural knee kinematics with single-plane fluoroscopy. J Biomech Eng 127(4): 692–699 Fujiwara T, Togashi K, Yamaoka T, Nakai A, Kido A, Nishio S, Yamamoto T, Kitagaki H, Fujii S (2004) Kinematics of the uterus: cine mode MR imaging. Radiogr Rev Publ Radiol Soc N Am Inc 24(1):e19. https://doi.org/10.1148/rg.e19 Glover GH, Pelc NJ (1988) A rapid-gated cine MRI technique. In: Magnetic resonance annual. Raven Press, New York, NY, pp 299–333 Im HS, Goltzer O, Sheehan FT (2015) The effective quadriceps and patellar tendon moment arms relative to the tibiofemoral finite helical axis. J Biomech 48(14):3737–3742. https://doi.org/ 10.1016/j.jbiomech.2015.04.003 Jensen ER, Morrow DA, Felmlee JP, Odegard GM, Kaufman KR (2015) Error analysis of cine phase contrast MRI velocity measurements used for strain calculation. J Biomech 48(1):95–103. https://doi.org/10.1016/j.jbiomech.2014.10.035 Jensen ER, Morrow DA, Felmlee JP, Murthy NS, Kaufman KR (2016) Characterization of three dimensional volumetric strain distribution during passive tension of the human tibialis anterior using cine phase contrast MRI. J Biomech 49(14):3430–3436. https://doi.org/10.1016/j. jbiomech.2016.09.002 Kaiser J, Bradford R, Johnson K, Wieben O, Thelen DG (2013) Measurement of tibiofemoral kinematics using highly accelerated 3D radial sampling. Magn Reson Med 69(5):1310–1316. https://doi.org/10.1002/mrm.24362 Kaiser J, Monawer A, Chaudhary R, Johnson KM, Wieben O, Kijowski R, Thelen DG (2016) Accuracy of model-based tracking of knee kinematics and cartilage contact measured by dynamic volumetric MRI. Med Eng Phys 38(10):1131–1135. https://doi.org/10.1016/j. medengphy.2016.06.016 Keegan J, Firmin D, Gatehouse P, Longmore D (1994) The application of breath hold phase velocity mapping techniques to the measurement of coronary artery blood flow velocity: phantom data and initial in vivo results. Magn Reson Med 31(5):526–536 Kinugasa R, Shin D, Yamauchi J, Mishra C, Hodgson JA, Edgerton VR, Sinha S (2008) Phasecontrast MRI reveals mechanical behavior of superficial and deep aponeuroses in human medial gastrocnemius during isometric contraction. J Appl Physiol (Bethesda, MD: 1985) 105(4): 1312–1320. https://doi.org/10.1152/japplphysiol.90440.2008 Lafortune MA, Cavanagh PR, Sommer HJ 3rd, Kalenak A (1994) Foot inversion-eversion and knee kinematics during walking. J Orthop Res 12(3):412–420. https://doi.org/10.1002/jor. 1100120314 Langner I, Fischer S, Eisenschenk A, Langner S (2015) Cine MRI: a new approach to the diagnosis of scapholunate dissociation. Skelet Radiol 44(8):1103–1110. https://doi.org/ 10.1007/s00256-015-2126-4 Manal K, McClay I, Stanhope S, Richards J, Galinat B (2000) Comparison of surface mounted markers and attachment methods in estimating tibial rotations during walking: an in vivo study. Gait Posture 11(1):38–45 Markl M, Alley MT, Pelc NJ (2003) Balanced phase-contrast steady-state free precession (PC-SSFP): a novel technique for velocity encoding by gradient inversion. Magn Reson Med 49(5):945–952. https://doi.org/10.1002/mrm.10451 Melchert UH, Schroder C, Brossmann J, Muhle C (1992) Motion-triggered cine MR imaging of active joint movement. Magn Reson Imaging 10(3):457–460 Moerman KM, Sprengers AM, Simms CK, Lamerichs RM, Stoker J, Nederveen AJ (2012) Validation of continuously tagged MRI for the measurement of dynamic 3D skeletal muscle tissue deformation. Med Phys 39(4):1793–1810. https://doi.org/10.1118/1.3685579 Moro-oka TA, Hamai S, Miura H, Shimoto T, Higaki H, Fregly BJ, Iwamoto Y, Banks SA (2007) Can magnetic resonance imaging-derived bone models be used for accurate motion measurement with single-plane three-dimensional shape registration? J Orthop Res 25(7):867–872. https://doi.org/10.1002/jor.20355
132
F.T. Sheehan and R.M. Smith
Pappas GP, Asakawa DS, Delp SL, Zajac FE, Drace JE (2002) Nonuniform shortening in the biceps brachii during elbow flexion. J Appl Physiol (Bethesda, MD: 1985) 92(6):2381–2389. https:// doi.org/10.1152/japplphysiol.00843.2001 Pelc NJ (1995) Flow quantification and analysis methods. Magn Reson Imaging Clin N Am 3(3):413–424 Pelc NJ, Bernstein MA, Shimakawa A, Glover GH (1991a) Encoding strategies for three-direction phase-contrast MR imaging of flow. J Magn Reson Imaging: JMRI 1(4):405–413 Pelc NJ, Herfkens RJ, Shimakawa A, Enzmann DR (1991b) Phase contrast cine magnetic resonance imaging. Magn Reson Q 7(4):229–254 Pelc NJ, Sommer FG, Li KC, Brosnan TJ, Herfkens RJ, Enzmann DR (1994) Quantitative magnetic resonance flow imaging. Magn Reson Q 10(3):125–147 Pelc NJ, Drangova M, Pelc LR, Zhu Y, Noll DC, Bowman BS, Herfkens RJ (1995) Tracking of cyclic motion with phase-contrast cine MR velocity data. J Magn Reson Imaging: JMRI 5(3):339–345 Pike GB, Meyer CH, Brosnan TJ, Pelc NJ (1994) Magnetic resonance velocity imaging using a fast spiral phase contrast sequence. Magn Reson Med 32(4):476–483 Regev GJ, Kim CW, Tomiya A, Lee YP, Ghofrani H, Garfin SR, Lieber RL, Ward SR (2011) Psoas muscle architectural design, in vivo sarcomere length range, and passive tensile properties support its role as a lumbar spine stabilizer. Spine 36(26):E1666–E1674. https://doi.org/ 10.1097/BRS.0b013e31821847b3 Riederer SJ (1993) Spatial encoding and image reconstruction. In: Bronskill MJ, Sprawls P (eds) The physics of MRI: 1992 AAP< summer school proceedings. American Instiutes of Physics, Woodbury, pp 135–165 Sheehan FT (2012) The 3D in vivo Achilles’ tendon moment arm, quantified during active muscle control and compared across sexes. J Biomech 45(2):225–230. https://doi.org/10.1016/j. jbiomech.2011.11.001 Sheehan FT, Drace JE (2000) Human patellar tendon strain. A noninvasive, in vivo study. Clin Orthop Relat Res 370:201–207 Sheehan FT, Zajac FE, Drace JE (1998) Using cine phase contrast magnetic resonance imaging to non-invasively study in vivo knee dynamics. J Biomech 31(1):21–26 Sheehan FT, Borotikar BS, Behnam AJ, Alter KE (2012) Alterations in in vivo knee joint kinematics following a femoral nerve branch block of the vastus medialis: implications for patellofemoral pain syndrome. Clin Biomech (Bristol, Avon) 27(6):525–531. https://doi.org/ 10.1016/j.clinbiomech.2011.12.012 Shibanuma N, Sheehan FT, Lipsky PE, Stanhope SJ (2004) Sensitivity of femoral orientation estimates to condylar surface and MR image plane location. J Magn Reson Imaging: JMRI 20(2):300–305. https://doi.org/10.1002/jmri.20106 Shibanuma N, Sheehan FT, Stanhope SJ (2005) Limb positioning is critical for defining patellofemoral alignment and femoral shape. Clin Orthop Relat Res 434:198–206 Shih YF, Bull AM, McGregor AH, Humphries K, Amis AA (2003) A technique for the measurement of patellar tracking during weight-bearing activities using ultrasound. Proc Inst Mech Eng H J Eng Med 217(6):449–457 Silder A, Reeder SB, Thelen DG (2010) The influence of prior hamstring injury on lengthening muscle tissue mechanics. J Biomech 43(12):2254–2260. https://doi.org/10.1016/j.jbiomech.2010.02.038 Sinha S, Shin DD, Hodgson JA, Kinugasa R, Edgerton VR (2012) Computer-controlled, MR-compatible foot-pedal device to study dynamics of the muscle tendon complex under isometric, concentric, and eccentric contractions. J Magn Reson Imaging: JMRI 36(2): 498–504. https://doi.org/10.1002/jmri.23617 Stone M, Davis EP, Douglas AS, NessAiver M, Gullapalli R, Levine WS, Lundberg A (2001) Modeling the motion of the internal tongue from tagged cine-MRI images. J Acoust Soc Am 109(6):2974–2982 Thompson RB, McVeigh ER (2004) Flow-gated phase-contrast MRI using radial acquisitions. Magn Reson Med 52(3):598–604. https://doi.org/10.1002/mrm.20187
3D Musculoskeletal Kinematics Using Dynamic MRI
133
Thunberg P, Karlsson M, Wigstrom L (2003) Accuracy and reproducibility in phase contrast imaging using SENSE. Magn Reson Med 50(5):1061–1068. https://doi.org/10.1002/ mrm.10634 van Kampen A, Huiskes R (1990) The three-dimensional tracking pattern of the human patella. J Orthop Res 8(3):372–382. https://doi.org/10.1002/jor.1100080309 Verbruggen SW, Loo JH, Hayat TT, Hajnal JV, Rutherford MA, Phillips AT, Nowlan NC (2016) Modeling the biomechanics of fetal movements. Biomech Model Mechanobiol 15(4): 995–1004. https://doi.org/10.1007/s10237-015-0738-1 Waterton JC, Jenkins JP, Zhu XP, Love HG, Isherwood I, Rowlands DJ (1985) Magnetic resonance (MR) cine imaging of the human heart. Br J Radiol 58(692):711–716. https://doi.org/10.1259/ 0007-1285-58-692-711 Wedeen VJ, Rosen BR, Chesler D, Brady TJ (1985) MR velocity imaging by phase display. J Comput Assist Tomogr 9(3):530–536 Wen H, Dou Z, Finni T, Havu M, Kang Z, Cheng S, Sipila S, Sinha S, Usenius JP, Cheng S (2008) Thigh muscle function in stroke patients revealed by velocity-encoded cine phase-contrast magnetic resonance imaging. Muscle Nerve 37(6):736–744. https://doi.org/10.1002/mus.20986 Westphal CJ, Schmitz A, Reeder SB, Thelen DG (2013) Load-dependent variations in knee kinematics measured with dynamic MRI. J Biomech 46(12):2045–2052. https://doi.org/ 10.1016/j.jbiomech.2013.05.027 Wilson NA, Sheehan FT (2009) Dynamic in vivo 3-dimensional moment arms of the individual quadriceps components. J Biomech 42(12):1891–1897. https://doi.org/10.1016/j.jbiomech. 2009.05.011 Wilson NA, Sheehan FT (2010) Dynamic in vivo quadriceps lines-of-action. J Biomech 43(11): 2106–2113. https://doi.org/10.1016/j.jbiomech.2010.04.002 Wood M, Wiang Q (1993) Motion artifacts and remedies. In: Bronskill MJ, Sprawls P (eds) The physics of MRI: 1992 AAP< summer school proceedings. American Instiutes of Physics, Woodbury, pp 383–411 Yamashita S, Isoda H, Hirano M, Takeda H, Inagawa S, Takehara Y, Alley MT, Markl M, Pelc NJ, Sakahara H (2007) Visualization of hemodynamics in intracranial arteries using time-resolved three-dimensional phase-contrast MRI. J Magn Reson Imaging: JMRI 25(3):473–478. https:// doi.org/10.1002/jmri.20828 Zhou H, Novotny JE (2007) Cine phase contrast MRI to measure continuum Lagrangian finite strain fields in contracting skeletal muscle. J Magn Reson Imaging: JMRI 25(1):175–184. https://doi.org/10.1002/jmri.20783 Zhu Y, Drangova M, Pelc NJ (1996) Fourier tracking of myocardial motion using cine-PC data. Magn Reson Med 35(4):471–480
Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion Richard M. Smith and Frances T. Sheehan
Abstract
Human movement is integral to daily life, it defines our species (the ability to walk upright and manipulate objects using an opposable thumb), and it is central to our ability to interact with our environment. As such, the study of human motion is dually important in our ability to optimize human functional ability. It provides a platform for understanding how pathology or injury affects human motion, so that we can both prevent and treat such pathologies. The earliest studies of human motion were mainly observational to qualify types of movements, while the current discipline and subdisciplines of human movement studies aim to quantify musculoskeletal kinematics, at times with submillimeter accuracy. The aim of this chapter is to discuss invasive and noninvasive methodologies for studying human motion with a focus on the reported accuracies, advantages, and limitations for each technique. Accuracies are presented throughout this chapter if they were reported as maximum average absolute or root mean squared errors for accuracy data for translational (in millimeters) and rotational data (in degrees) in order to simplify the reporting of cumulative accuracies from relevant articles. Thus, this review will highlight the current state of each methodology, as a platform for future investigators to build on these technologies. Keywords
Validation • Accuracy • Magnetic resonance imaging • MRI • Cine MRI • Cine phase contrast • CPC motion capture • Fluoroscopy • Single-plane videoradiography • Biplane videoradiography • Ultrasound • Muscle • Skeletal • R.M. Smith • F.T. Sheehan (*) Rehabilitation Medicine Department, Functional and Applied Biomechanics Section, National Institutes of Health, Bethesda, MD, USA e-mail: [email protected]; [email protected] # This is a U.S. Government work and not under copyright protection in the US; foreign copyright protection may apply 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_194
135
136
R.M. Smith and F.T. Sheehan
Musculoskeletal • Computed Tomography • CT • Motion capture • Optoelectronic tracking system • OTS • Pose estimation
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Invasive Methods in Motion Analysis and Radiographic-Based Studies . . . . . . . . . . . . . . . . . . . . . . Motion Analysis with X-Ray Fluoroscopy and Computed Tomography . . . . . . . . . . . . . . . . . . . . . . X-ray Stereophotogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fluoroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Computed Tomography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Noninvasive Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Motion Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic Resonance Imaging (MRI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
136 137 137 139 139 139 142 143 143 145 146 149 151
Introduction Human movement is integral to daily life. It defines our species (the ability to walk upright and manipulate objects using an opposable thumb) and is central to our ability to interact with our environment. As such, the study of musculoskeletal kinematics is crucial as it (1) provides qualitative and quantitative information about the musculoskeletal system, (2) aids our understanding of how pathology or injury affects human motion, (3) increases our ability to prevent and treat such pathologies, and (4) is a basis from which to optimize human functional performance. The aim of this chapter is to discuss invasive and noninvasive methodologies for estimating the static and dynamic joint pose (position and orientation) and muscle dynamics. There is a focus on the reported accuracies, advantages, limitations, as well as future directions for each technique. This review does not intend to summarize every published work in the field. However, great care was taken to perform a robust search of both the PubMed and EMBASE databases for any article on human musculoskeletal kinematics. Both authors systematically sorted the resulting references by title, abstract, and full text to include only those dealing directly with accuracy of musculoskeletal tracking techniques. For the purpose of this review, invasive techniques will refer to methods that introduce a break in skin and surface barriers to insert research instruments, such as bone-anchored pins or screws and radiopaque beads. Accuracies will be presented throughout the chapter as maximum average absolute or root mean squared (RMS) errors for translational (in millimeters) and rotational data (in degrees) to simplify discussion of the cumulative accuracies from relevant articles. This decision was made, as average error represents the bias of a measurement technique. Bias only defines accuracy when numerous measures (at least 30–50) of the same quantity can be acquired and then averaged to remove
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
137
the inaccuracy from random error (Materials 2010). Additionally, this chapter seeks to distinguish the technologies which provide static two-dimensional analyses from others which provide truly three-dimensional assessments of musculoskeletal motion. Thus, this review will highlight the current state of each methodology as a platform for future investigators to build on these technologies.
State of the Art Historical studies of human motion date back through the millennia and may predate the ancient Greek philosophers, Hippocrates, Plato, and Aristotle (Abernethy 2013). However, the modern study of human movement evolved as an academic discipline in the mid-twentieth century with photographic and electrogoniometric techniques utilizing both in vivo and cadaver studies (Abernethy 2013). Early noninvasive gait studies were followed by more invasive methods (e.g., bone-anchored pins) as researchers sought greater accuracy in recording the underlying skeletal motion. In the following decades, fluoroscopy and computed tomography (CT) were adapted with and without invasive methods to study skeletal kinematics with the drawback of ionizing radiation exposure. Yet, advances in X-ray-based modalities have led to decreases in radiation exposure. Currently in the research community, there has been a refocus to noninvasive, nonionizing methods to study musculoskeletal dynamics. These include magnetic resonance imaging (MRI), ultrasonography, and motion capture systems. Both MRI and ultrasound provide the added benefit of being capable of tracking muscle motion. It is these non-irradiating, noninvasive techniques which have paved the way for greater research applicability especially in voluntary research studies involving vulnerable populations (e.g., children and pregnant women). From a global perspective, the earliest studies were based on observations to qualify the types of human movements, while the current discipline and subdisciplines of human movement studies aim to quantify musculoskeletal kinematics, at times with submillimeter accuracy. It is this current aim of human motion studies that makes it a truly evolving field. For details on the historical aspects of human motion studies, see the chapter on ▶ “Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries.”
Invasive Methods in Motion Analysis and Radiographic-Based Studies Highly invasive techniques with percutaneous skeletal markers and bone-anchored screws, intracortical pins, and radiopaque beads were adapted from earlier motion capture studies as a means of investigating skeletal kinematics without the effects of soft tissue artifacts. In the study of human kinematics, invasive procedures are far reaching given their adaptability to many, if not all, imaging modalities. Invasive
138
R.M. Smith and F.T. Sheehan
methodologies have been used to study the hip and pelvis (Holden et al. 1997; Neptune and Hull 1995) and both upper and lower appendages, especially the knee joint, ankle complex, and glenohumeral joints (Dal Maso et al. 2014; Lafortune 1984; Manal et al. 2000; Reinschmidt et al. 1997). Although intracortical pins are limited in their applicability to in vivo human studies, due to their invasiveness; they have become one of the gold standards for validating other measurement techniques. Initial work using photogrammetry coupled with intracortical pins by Levens et al. (1948) and later by Lafortune (1984) provided insights into multibody pose estimation during the gait cycle. Early validation studies focused on quantifying the accuracy of pins compared to surface trackers. Over time, intracortical pins became their own gold standard. Employing both photographic and radiographic methods, Lafortune et al. (1992, 1994) ascertained an average maximal detectable difference of 0.4 mm radiographically and 0.5 mm for their photogrammetric system calibration. Later photogrammetric work with intracortical pins showed that pins were superior to external markers when studying tibiofemoral rotations but not superior to surface markers for rotations of the tibiocalcaneal joint (Reinschmidt et al. 1997). More recent studies of the upper extremity with intracortical pins and computed tomography (CT) have demonstrated RMS error of 0.15 mm and 0.2 for glenohumeral kinematics using a computer-based filtering technique (Dal Maso et al. 2014, 2016). Besides their invasiveness, the other drawback to using intracortical pins is the potential for the pins to loosen or dislodge. In an effort to mitigate errors related to soft tissue artifact from skin-surface markers and to avoid the invasiveness of bone-mounted markers, Holden et al. (1997) developed and validated a percutaneous skeletal tracker, a clamp-like device attached to the periosteum (rather than cortex) of the underlying bone via halo pins (2 mm diameter). In a preliminary study, using unfiltered data, an absolute error of 1.7 mm was reported (Holden et al. 1997). Percutaneous trackers offer an advantage over intracortical pins because their construct provides a compressive force to minimize soft tissue movement. Percutaneous skeletal trackers are often considered “minimally invasive” when compared to larger and more depth-intrusive intracortical pins. These trackers allow for accurate analyses, on par with intracortical pin studies of (Dal Maso et al. 2014). Invasive methodologies coupled to ionizing radiation imaging technologies (Brainerd et al. 2010; Tashman and Anderst 2003; Veress et al. 1979; You et al. 2001) and nonionizing MRI studies (Pelc et al. 1995) have also been used to study motion. Radiopaque bone markers, such as lead, steel, or tantalum beads, have been used in vitro and in vivo studies with fluoroscopic modalities in a technique widely known as radiostereometric analysis. However, in vivo accuracy can be limited by an inability to visualize all markers, due to field of view (FOV) limitations or other anatomic structures blocking the view of the implants (e.g., the contralateral limb). Overall, invasive studies have been used as the gold standard for validation of other measurement techniques, due to their inherent accuracy. However, there are several disadvantages to their use, such as risk of infection, bleeding, and the use of anesthesia affecting the “natural” motion of the test subject. These limit their use for large-scale population studies, studies of pathology, and longitudinal assessments. In
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
139
summary, invasive procedures have lost prestige in human research due to their low favorability in recruiting subjects for non-medically necessary research, especially among minors. Notwithstanding, invasive studies continue to direct veterinary studies, as well as cadaveric studies.
Motion Analysis with X-Ray Fluoroscopy and Computed Tomography X-ray Stereophotogrammetry The first sets of experiments with X-ray stereophotogrammetry provided two-dimensional information of relative bone pose and movement (Lippert et al. 1975; Tranberg et al. 2011; Veress et al. 1979). These early systems were limited by poor resolution, motion blurring, and artifacts due to magnification and perspective. Over time, digitization procedures, using bone-specific coordinate transformations, were adopted to correct for such distortions. A major disadvantage with these two-dimensional techniques was the constraints on what movements could be studied; for example, studies of the lower limbs were limited to small incremental angular movements between poses.
Fluoroscopy Fluoroscopy evolved from the need for a noninvasive method to track joint function, particularly of the spine and knee. Fluoroscopy exposes subjects to continuous ionizing X-rays which can then be recorded to provide looping video images. Hence, singleplane and biplane videoradiography are adaptations of this X-ray technique. In contrast to earlier plain film X-ray, registration of dynamic fluoroscopic images to 3D bone models enables dynamic noninvasive tracking of skeletal kinematics. In general, these techniques began with a library-based imaging matching methodology presented by Banks and Hodge (1996) for single-plane fluoroscopy. With increased access to computational power, the library-based technique gave way to the current 2D to 3D matching methodologies for both single-plane and biplane videoradiography. The advantages of fluoroscopy are its ability to noninvasively assess joint function under dynamic activities and, in the case of biplane videoradiography, its submillimeter accuracies. For single-plane fluoroscopy registered to CT-based bone models, the maximum in-plane errors across studies range from 1.6 to 2.0 mm and 1.0 to 1.6 (Table 1). The out-of-plane errors tend to be higher, and for this reason, single-plane fluoroscopy has been primarily recommended for studies evaluating planar motion (Fregly et al. 2005). Biplane videoradiography can provide submillimeter accuracies for tracking 3D motion (Table 2). One of the most realistic validation studies was based on comparing the in vivo tracking of implanted tantalum beads to quantify tibiofemoral motion during running in three human subjects (Anderst et al. 2009). The maximum reported RMS error for
140
R.M. Smith and F.T. Sheehan
Table 1 Single-plane videoradiography accuracies. Studies were included if a root mean square (RMS) or average absolute (AA) error was available Study
Year
Object imaged
Motion
San Juan
2010
Harvested cadaver
1D translation
Lina Cerciello Tang
2013 2011 2004
Fernandezb
2008
Humerus and scapula Cadaver spine Dried vertebrae Cadaver knee (PF joint) Dried tibial and femoral bones
Maximum RMS or ave. abs error Translational Rotational IP OP IP OP 0.43 mm
Static Static Static
0.3 mm 2.0 mm 1.6 mm
2.5 mm
Static
2.0 mm
1.7 mm
0.5 2.1
2.9
3.8 mm
1.0
1.9
Abbreviations: IP in-plane, OP out -of -plane, 1D one-dimensional, PF patellofemoral a Estimated from Fig. 10 in (Lin et al. 2013) b Based on a single pose
tracking tibiofemoral motion was 1.5 mm and 1.8 . For individual bones, these RMS errors dropped to a maximum of 0.9 mm. For all fluoroscopic techniques, the capture FOV is small, limiting the ranges and types of motion that can be studied. In one study (Wang et al. 2015), heel strike and toe off needed to be captured separately, as the length of the foot could not be captured within the available FOV. A recent study by Guan et al. (2016) presents a system that partially overcomes this limitation by mounting a fluoroscopy system onto a computer-controlled sliding track that follows a subject during a walk, or potentially a run, along a linear track. These authors reported excellent accuracies (0.8 mm and 0.8 , max RMS error), yet the simulated walking speed was approximately 50% slower than typical (Bohannon and Williams Andrews 2011). Although rarely discussed, another limitation of fluoroscopy studies is the data processing times. One study reported analysis times up to 12.5 h (Ohnishi et al. 2010). Longer processing times make the applicability of biplane videoradiography in clinical, non-research, settings less likely. Moving away from CT models to MR-based models, which typically require lengthy manual segmentation of the bones of interest, will serve to increase computation time, whereas ever-increasing computer processor speeds will serve to reduce computation time. For researchers looking to apply this modality, care must be given in terms of radiation exposure and the true accuracy of the system being used. The capture rate is system dependent, with the majority of commercially available systems having capture rates of just 30 frames/s. This limits studies to motion at unnaturally slow speeds (Tashman 2008). Several custom-built videoradiography systems with capture rates of 200 frames/s and 250 frames/s (Anderst et al. 2009; Guan et al. 2016)
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
141
Table 2 Biplane videoradiography accuracies. Studies were included if a root mean square (RMS) or average absolute (AA) error was available. Note, the Anderst et al. (2009) and the Ohnishi study are listed three and two times, respectively, to highlight three and two unique validations done within these studies
Study You
Year 2001
deBruina
2008
Beyb
2006
Beyb
2008
Wangc
2008
Object imaged In vivo canine tibia Sawbone scapula Cadaver – humerus/ scapula Cadaver – PF Ovine spine
Anderst
2009
In vivo TF
Anderst
2009
In vivo tibia
Anderst Ohnishi Ohnishi Wang
2009 2010 2010 2015
In vivo TF Ovine knee Ovine knee Cadaver rear foot
Thorhauerd Guan
2015 2015
Cadaver TF Cadaver TF
Frame rate
Shutter speed
Motion Walking (1.5 m/s) Static
(Frames/s) 250
(μs) 0.5
Dynamic (unknown)
50
Dynamic (unknown) Translation (0.017 m/s) Running (2.5 m/s) Running (2.5 m/s) Static Static Static Simulated walking (1.0 m/s) Static Simulated walking (0.7 m/s)
60
Maximum RMS or AA error Trans (mm) Rot 0.8 3.9 0.3
0.4 a
2.0
0.4
0.9
2.0
0.4
0.9
30
0.2
250
0.5
1.5
250
0.5
0.9
250 30 30 100
0.5
1.0
0.26 0.5 0.4 0.7
250 200
0.5 5.0
1.5 0.8
1.8
0.9 0.7 0.6
0.8
Abbreviations: TF tibiofemoral a RMS error calculated from Table 2 in text b Motion rate not provided c Translation was a 1 degree of freedom movement and error increased to 0.4 mm when an MRI model was used d Accuracy listed for locating point of contact
have enabled biplane videoradiography to capture movement at realistic speeds. When these fast frames rates are coupled with fast shutter speeds, most issues associated with blurring are removed (Tashman 2008). Yet, based on numerous studies, quantifying dynamic joint kinematics is less accurate than quantifying a static pose, even with customized systems (Anderst et al. 2009; Ohnishi et al. 2010). The system parameters, the speed of motion, the shape/size of the bone being tracked, and the 3D model used can all affect tracking accuracy. For example,
142
R.M. Smith and F.T. Sheehan
tracking single bones is more accurate than tracking two bones relative to each other (Anderst et al. 2009; Ohnishi et al. 2010). Further, using an MRI model (de Bruin et al. 2008; Thorhauer and Tashman 2015; Wang et al. 2008) or a lower-resolution CT model (Fox et al. 2011) can increase errors. One key issue raised by Anderst et al. (2009) is that capturing dynamic bone kinematics versus static pose is less accurate. Thus, researchers must carefully discern if previous validation studies are truly applicable to their current experimental design or if a validation study for their individual study is needed. For example, two studies (Bey et al. 2006, 2008) reported excellent accuracies for tracking patellofemoral and humeral motion, yet the motion rate evaluated was not provided. Thus, it is unclear how to apply their results to future studies. Both single-plane and biplane videoradiography have advanced the study of skeletal kinematics. Single-plane fluoroscopy has its strengths for studies involving in-plane motion (Fregly et al. 2005; Cerciello et al. 2011; Fernandez et al. 2008), as such it is an indispensable clinical tool during intraoperative planning and postsurgical follow-up. Biplane videoradiography is a continually evolving technique which can provide submillimeter accuracies. Additionally, biplane shows great adaptability in its application to the study of functional tasks, such as walking and running (Anderst et al. 2009; Guan et al. 2016). Innovations in its use, such as creating a mobile videoradiography platform (Guan et al. 2016), allow for a wide variety of joints and joint range of motions to be captured with greater accuracy than motion capture systems, discussed later in this chapter. For details on biplane videoradiography, see the chapter on ▶ “Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography.”
Dynamic Computed Tomography Computed tomography (CT) is another common noninvasive clinical tool which uses ionizing radiation to reconstruct three-dimensional volumetric anatomical data from planar datasets. This technology has also progressed from static CT to both contrast-enhanced CT and four-dimensional (dynamic) CT. Similar to X-ray stereophotogrammetry, early studies in conventional CT (Dupuy et al. 1997; Fischer et al. 2001; Lee et al. 2014; Rogers et al. 2005; Schutzer et al. 1986; Zuhlke et al. 2009) provided information about static alignment for various joints. In one study, researchers utilized CT and electrogoniometry to study joint pose in 6 degrees of freedom with results validated up 1.2 mm and 0.5 ; RMS error (Jan et al. 2002). CT studies in rigid body pose estimation and registration block methodologies (Fischer et al. 2001) paved the way for future dynamic CT studies. For example, Buffi et al. (2013) utilized static datasets of relative CT bone poses to create manual computerbased bone models of the carpometacarpal joints of the hand. This in vivo study warrants commendation as it provided data on in vivo hand range of motion fitted to a biomechanical model of hand kinematics. These authors reported absolute error between the CT and digitized models up to 6.2 for supination pronation, up to 1.2 for flexion-extension, and up to 1.2 in abduction-adduction (Buffi et al. 2013).
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
143
Currently, there are seemingly few validation studies for dynamic CT on skeletal kinematics. Of the studies that have focused on validation, numerous authors do not report accuracies in RMS or absolute errors, but rather they report bias or precision measurements (Gondim Teixeira et al. 2017; Kerkhof et al. 2016; Goto et al. 2014; Zuhlke et al. 2009). One validation study reported RMS localization errors between the dynamic and static CT data in the range of 0.023–0.139 mm (Zhao et al. 2015). Dynamic CT does have great clinical applicability and has already been used to study multiple joints (Dupuy et al. 1997; Kalia et al. 2009; Williams et al. 2016). Dynamic CT, like other X-ray technologies, is somewhat limited in its use due to ionizing radiation exposure. In addition, due to public perception and fears about prolonged radiation exposure, CT studies are further limited for use in underrepresented research populations, such as children and pregnant women or in anatomical regions close to the thyroids or genitals. Technical limitations of CT include capture FOV, data acquisition speed, and motion artifacts. Although the bore of CT scanners has increased in recent years to accommodate more complex positioning and subjects with increased body habitus (80–90 cm in most commercial scanners), the capture FOV for many systems is around 50–60 cm, which does not allow the full range of motion for many joints to be studied. Motion artifacts and data acquisition time will continue to improve with spiral (or helical), cone-beam, and multi-detector CT technologies. However, faster CT scan times are proportional to greater radiation exposure (Biswas et al. 2009).
Noninvasive Imaging Motion Capture Stereophotogrammetry is the science of utilizing dynamic video images, static photography, or radiography to quantify human motions. Photogrammetric studies of gait and motion date back to the late 1800s and are most attributed to the work of E. Muybridge, with his work on animal and human motion studies using freeze frame photographic analysis (Cappozzo and Paul 1998). This work in freeze frame photography was later advanced through a collaboration with E.J. Marey. Early studies (Levens et al. 1948; Eberhart and Inman 1951) in human motion capture involved a system of high-speed cameras and skin-surface markers. These skinmounted targets often consisted of retroreflective (passive) or light-emitting (active) markers clustered on the skin directly above a bone or other anatomical landmarks. In one experimental design, participants were photographically recorded walking along glass walkways to estimate human pose during gait, with an estimated accuracy of approximately 13 mm (Eberhart and Inman 1951). Progressively, researchers identified that the movement and deformation of the underlying tissue (e.g., adipose, muscle) caused movement of the skin-mounted markers, relative to the underlying bones. This affected the accuracy of tracking bone, with errors up to 40 mm, depending on the location of trackers and the activity studied (Karlsson and
144
R.M. Smith and F.T. Sheehan
Lundberg 1994; Cappozzo et al. 1996). As such, the previously discussed invasive methodologies (pins, beads, percutaneous trackers) were utilized as an attempt to directly measure and track anatomical points of interests. Initial studies of the knee with 2D goniometric systems reported average absolute errors up to 2 based on the relative placement of the device in relation to the central axis of the joint (Kettelkamp et al. 1970). Although goniometry showed consistent cyclogram patterns for certain linear activities such as walking, running, and knee flexion-extension tasks (Townsend et al. 1977); it was limited to angular motions, and the test-retest reliability depends on the correct placement of the device (Chao 1980). Modern motion capture combines multiple cameras (infrared or high-speed video-capture), calibration hardware and software, and multiprocessor computer systems to measure whole-body motion. These systems can also be tuned to evaluated small regions (e.g., facial motion). Optoelectronic tracking system (OTS) is one form of motion capture technology that utilizes this collection of hardware and software. Most OTS setups require a large open space, wherein the cameras can be oriented according to manufacturer’s specifications. This open space requirement results in motion capture being the one measurement system that can look at the largest variety of functional tasks (e.g., walking, running, pitching, golf swing, etc.). Submillimeter accuracies in locating the skin-mounted markers are achievable, but the actual accuracies for tracking skeletal kinematics using OTS are limited by lab setup (camera-subject operating range and software calibration) and anthropometric subject variables. The biggest issue in terms of the accuracy is the mismatch between the underlying bone motion and the motion tracked by skin markers. Although numerous algorithms (Wan and Nelson 2001; Todorov 2007) have been developed to compensate for this mismatch, errors attributed to skin marker artifact have a reported mean absolute error range of 0.5–7.7 mm (Chiari et al. 2005). Overall, probabilistic approaches to pose estimation have shown promise at reducing these errors (Todorov 2007). One study (Wilson et al. 2009) compensated for skin marker artifacts in motion capture by creating a thermoplast cup for the patella. By doing so, this study demonstrated the feasibility of tracking patellofemoral motion during a specific exercise (slow squat) using OTS. The overall accuracies were reported at 1.2 / 1.1 mm (average absolute errors). The term “motion capture” envelops more than just OTS systems. Electromagnetic tracking systems evaluate 3D positions relative to a magnetic field but are limited in their accuracy by artifacts created by various metals in the workspace (Milne et al. 1996; Meskers et al. 1999). Force platforms record the interaction forces between the body and the ground. These are critical data for inverse dynamics analyses of human motion, which can provide estimates of muscle forces and neuromuscular control strategies. Lastly, current advances in wearable technologies are fostering the development of portable magneto-inertial units (MIMU) that feature accelerometers, gyroscope, and magnetometers for measuring joint kinematics. Wearable sensors have been tested against known clinical tests such as “timed up and go” tasks (Salarian et al. 2010) and commercial OTS systems (Simoes 2011).
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
145
MIMUs are currently limited by the accuracy of how the device estimates its pose relative to assigned reference frames (Cereatti et al. 2015). However, this technology shows great promise for future studies of multi-segment rigid body kinematics and the parameters of pathological gait in both clinical and research settings (Cereatti et al. 2015; Crabolu et al. 2016). Overall, motion capture analysis excels in its noninvasive nature and the ease at which it can quantify multi-joint movements. However, the results of such estimations have a relatively lower accuracy in tracking skeletal motion, compared to biplane videoradiography and dynamic MRI. An advantage, as well as a disadvantage, of OTS systems is the large quantity of output variables it can provide. Although this wealth of data can provide crucial insights into human movement, it can be difficult to extract the key variables that are relevant to a specific question. OTS and electromagnetic trackers must be calibrated on a case-by-case basis to ensure accurate analyses. Looking forward, small wearable activity monitors are popular consumer electronics for tracking general activity levels. In addition, there is an emerging market for powerful inertial trackers, which are currently being validated to track real-time gait measurements. Further details on motion capture systems including reflective markers or markerless systems can be found in the chapter(s) on: ▶ “Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis,” ▶ “3D Dynamic Pose Estimation from Marker-Based Optical Data,” ▶ “3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers,” ▶ “3D Dynamic Pose Estimation from Markerless Optical Data,” ▶ “Physics-Based Models for Human Gait Analysis,” ▶ “Clinical Gait Assessment by Video Observation and 2D Techniques.” For further details on inertial trackers and wearable sensors, please refer to book chapter(s) on: ▶ “Gait Parameters Estimated Using Inertial Measurement Units,” ▶ “Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units,” ▶ “Induced Acceleration and Power Analyses of Human Motion.”
Dynamic Ultrasound Diagnostic ultrasonography offers noninvasive, nonionizing imaging of the musculoskeletal system. It utilizes high frequency sound waves to produce a grayscale display based on tissue density and flow properties. Rigid or semirigid anatomic structures such as bones, blood vessels, and soft tissue reflect the transmission of acoustic waves causing propagated vibrations, which are in turn interpreted by a mechanical sensor to generate real-time dynamic images. There are several imaging protocols for diagnostic ultrasound such as Doppler (D), brightness (B), and motion (M) modes, among others. The mode and frequency of the selected ultrasound protocol depend on the depth and size of the region or structure imaged. Major advantages of ultrasound are its wide availability and its portability, which has fostered extensive use of ultrasound in both research and clinical settings. In addition, it is adaptable to studying a broad range of musculoskeletal structures
146
R.M. Smith and F.T. Sheehan
and movements, and it offers a wide field of study which allows the probe to be custom fitted to various body segments. Musculoskeletal sonography is capable of measuring structures at both the macroscopic and gross tissue level. Ultrasound technologies can track single muscle fascicle motion and tendon velocities (Tat et al. 2015). Ultrasound has also been coupled with surface or fine wire electromyography to study muscle fiber activations (Byrne et al. 2005; Pulkovski et al. 2008). For gross structure examinations, dynamic ultrasonography has wide applications in studying hyoid bone motion, pelvic floor musculature contraction (Braekken et al. 2008), ankle motion (Telfer et al. 2014), and patellar tracking (Shih et al. 2003). Static ultrasound has been coupled with motion capture and computer-assisted algorithms to study hip kinematics (Jia et al. 2016). Jia et al. (2016) produced faster processing times with their automated registration pipeline and reported maximum absolute errors up to 0.2 mm and 4 , for segmentation and registration, respectively. A major disadvantage of ultrasound is its operator bias, as data quality is dependent on the operator’s level of training and experience. As such, many ultrasound studies aim to standardize their results with measures of class or rater reliability. Shih et al. (2003) performed an in vivo dynamic study of medial-lateral patellar displacement during sitting and squatting using a custom knee brace fitted with a lateral mounted ultrasound transducer. These authors validated their technique against a static interventional MRI and even provided follow-up studies that showed increased accuracies, which were reported as bias (Shih et al. 2003, 2004). Current limitations of ultrasonography include image resolution, which is both mode and orientation dependent, its inability to capture data if an air interface is present, and its relative difficulty in studying deep musculoskeletal structures or ligaments and joints. As a whole, ultrasound offers great potential for noninvasive musculoskeletal studies involving tendons, ligaments, and muscle. For further discussion on ultrasound use in musculoskeletal kinematics, see the chapter on ▶ “Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and Ligament.”
Magnetic Resonance Imaging (MRI) MRI uses the magnetic properties of hydrogen atoms in tissue to create 3D anatomical images of bone, soft tissue, and cartilage, as well as other structures. MRI is a noninvasive and nonionizing technique that provides three-dimensional, quantitative data with high-resolution and accuracy. High-resolution static MRI sequences (e.g., gradient echo, fat suppression, proton density weighted) can be used to capture anatomic images of joints in various poses. Static MRI does not allow for full dynamic analysis throughout the motion cycle. However, the advantage of this process is that multiple parallel scans can be segmented to provide 3D musculoskeletal models that have become the foundation for dynamic musculoskeletal models at the single and multiple joint levels. Static MRI has provided the ability to measure such quantities as musculotendon and ligament paths, muscle fascicle direction and length,
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
147
tendon/ligament moment arms, relative muscle strength (in terms of muscle volume), bone shape, cartilage morphology, and numerous other properties. In addition, static MRI has been used to evaluate joint pose during active muscle activation, joint pose during weight bearing (Boden et al. 2009), and muscle volume as it relates to human movement (Eng et al. 2007; Im et al. 2014). In terms of evaluating musculoskeletal kinematics, there are three primary MRI techniques: cine MRI (Brossmann et al. 1993; Borotikar et al. 2012; Kaiser et al. 2016), MRI tagging (Pipe et al. 1991; Niitsu et al. 1992), and cine phase-contrast (CPC) MRI (Finni et al. 2006; Jensen et al. 2015; Pappas et al. 2002; Sheehan et al. 1998). CPC MRI is the only modality that can noninvasively track 3D in vivo skeletal and muscle movement (Table 3). It has the further advantage that it does Table 3 MRI Accuracies. Studies were included if a root mean square (RMS) or average absolute (AA) error was available. The table is broken into three sections. The top three studies focus primarily on muscle/soft tissue. The next three focus primarily on quantify skeletal kinematics by registering a low- to a high-resolution mode. The last three are focused primarily on using cine phase contrast (CPC) to measure skeletal motion
Study Drace
Year 1994
Pelc
1994
Moerman
2012
Object imaged Muscle (ex vivo bovine) In vivo canine heart Gelatin phantom
Fellowsa
2005
Cadaver PF joint
Borotikarb
2012
Kaiserb
2016
Sheehan
1998
Barrancec
2005
Behnam
2011
In vivo human PF joint Femoral and tibial bone segments Human cadaver bone and gelatin phantom Ex vivo bovine tibiofemoral Phantom
Motion Dynamic
Modality CPC
Dynamic
CPC
Linear depression Static
MRI tagging Low to high res Cine to model Cine to model
Dynamic Dynamic
System 1.5 T GE 1.5 T GE 3.0 T Philips 1.5 T GE 3.0 T Philips 3.0 T GE
Dynamic
CPC
1.5 T GE
Dynamic
CPC
Dynamic
CPC
1.5 T GE 3.0 T Philips
Maximum RMS or AA error Trans (mm) Rot 1.0 1.32 1.02 0.9
1.8
0.9
1.7
0.60
1.1
0.9
3.67 0.3
1.0
Abbreviations: bcine to model implies using a low spatial resolution dynamic capture and defining kinematics by registering the low-resolution images to the high-resolution images. Note two different types of cine imaging were used in these studies a Low to high res: A low-resolution static model was fitted to a high-resolution model b Cine to model: A multiplane cine model was fitted to a high-resolution model c Only a single degree of freedom was validated
148
R.M. Smith and F.T. Sheehan
not expose subjects to ionizing radiation. Yet, dynamic MRI is limited by the confined space within a closed-bore environment, which reduces the joint motions that can be studied, and it typically requires subjects to perform repetitive motion during data capture. Cine MRI was the first of the dynamic MRI techniques to be applied to the study of quantitative joint kinematics, with a focus on the temporomandibular (Burnett et al. 1987; Maniere-Ezvan et al. 1999) and patellofemoral (Brossmann et al. 1993; Ward et al. 2002) joints. This technique produces a series of anatomic images depicting cyclic motion by synchronizing the data collection to a repeated motion cycle. Recently, multiplane cine (MPC) MRI, which is cine MRI that captures multiple anatomic planes of data at each time point, has been used to track joint kinematics for the patellofemoral and tibiofemoral joints and cartilage contact for the patellofemoral joint, with submillimeter accuracy (Table 3). This was accomplished by creating spatially sparse 3D models (at each time point) of the bones being evaluated from MPC data (Borotikar et al. 2012; Kaiser et al. 2016). Independently fitting each of these temporal models to a static high-spatial-resolution 3D model of the same bone provided a transformation matrix that defined the required translation and rotation for transitioning the bone from the dynamic time frame to the static state. By acquiring the transformation matrices for all dynamic time frames to the static state, the kinematics throughout the motion were defined with submillimeter accuracy. MRI tagging enables muscle strain to be quantified by automatically tracking a signal void grid, which is established at the start of data capture. In short, this process destroys the MR signal of the tissue within the imaging plane in an even grid pattern (Pipe et al. 1991; Niitsu et al. 1992). This grid is inherent to the tissue; thus as the muscle deforms, the grid deforms with it. Visually, the image appears as a standard anatomical image with a black grid overlaid on it. Thus, tracking the grid points allows for tracking of muscle deformation with accuracies on the order of a millimeter (Table 3). Although many of the validation studies (Pipe et al. 1991; Niitsu et al. 1992; Sprengers et al. 2013) demonstrated the feasibility of using this technique to track skeletal muscle deformation, tagging has been primarily used in the arena of quantifying cardiac dynamic properties. Cine phase contrast (CPC) combines the properties of both PC and cine MRI sequences to allow cyclical anatomical and velocity data in three orthogonal directions (vx, vy, and vz) to be acquired. In total, these data provide images of the moving anatomy throughout the movement, along with the 3D velocity of each pixel within the imaging plane. Integrating the velocity data (Pelc et al. 1995) allows for tissues within the imaging plane to be tracked three-dimensionally over time with submillimeter accuracies. In doing so, dynamic musculotendon paths, moment arms, tendon strain, 3D joint kinematics, and muscle strains can be tracked throughout the motion. Unique to MRI, this modality captures dynamic data and static poses using similar, but distinctive, techniques. Since integration of CPC is used to determine the future position of an anatomical point, tracking skeletal kinematics using CPC (Behnam et al. 2011) has better accuracy than tracking joint pose using static MR (Fellows et al. 2005). In its earliest application, CPC showed high
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
149
accuracy in tracking skeletal muscle up to 1 mm, ex vivo (Drace and Pelc 1994), and up to 1.32 mm, in vivo. With improvements in MR hardware and software design, along with increases in magnetic strength (1.5 to 3.0 Tesla), increased signal-to-noise (SNR) ratios, improved temporal resolution, and reductions in the number of required motion rates were achieved. Behnam et al. (2011) demonstrated an accuracy of 0.3 mm for tracking the 3D motion of a phantom using the more advanced 3 T MRI scanner. Although CPC can track motion in 3D, it is limited in its ability to quantify 3D volumetric data due to imaging time constraints. Thus, CPC has also been combined with the MPC-to-modeling methodology to provide an accurate method for tracking joint contact with an average absolute error of 0.9 mm/1.7 (Borotikar et al. 2012). In comparison, Fellows et al. (2005) registered a low-resolution static MRI model to a high-resolution static model to evaluate loaded knee pose at various flexion angles. They reported an accuracy of 0.9 mm/1.8 (Fellows et al. 2005). Overall, dynamic MRI sequences offer a safe, nonionizing method to quantify in vivo 3D musculoskeletal kinematics. While long acquisition times have been cited in past studies, MRI sequences, such as real-time MRI (Asakawa et al. 2003) and fast-PC MRI sequences, continue to advance the field by drastically reducing the scanning time and the required number of motion cycles. Current closed-bore MRI systems limit the field of study and the types of dynamic motions (such as weightbearing exercises) that can be evaluated. Open-bore systems exist but are not widely used due to lower resolution and reduced signal-to-noise SNR. Current progress in MRI technology shows promise for continued improvements in temporal resolution, imaging time, and SNR. For greater detail on dynamic MRI theory and applications, refer to the chapter on ▶ “3D Musculoskeletal Kinematics Using Dynamic MRI.”
Future Directions In total, these various tools for evaluating in vivo musculoskeletal kinematics have greatly enhanced our understanding of the function and dysfunction, in the case of injury and pathology, of the neuromuscular and musculoskeletal systems. Modern research is grounded in the principle of minimizing harm to subjects. Thus, there should be a fine balance between the use of invasive or otherwise harmful techniques and the intended research goal of advancing scientific knowledge. This is not to say that invasive techniques do not have a place. For example, tracking a total knee replacement (TKR) using single-plane fluoroscopy has been proven to be quite accurate (Banks and Hodge 1996), but it is inherently an invasive technique, based on the need for an implanted TKR. Yet, as the decision for surgery is a clinical one, using fluoroscopy post-surgery is a noninvasive, inexpensive, and expedient method for tracking the implant functionality over time. Next, the most accurate technique is not always the best technique for every question. For example, a sparsely accurate pedometer can be a perfect tool for evaluating a subject’s general level of activity, whereas to understand changes in muscle control, motion capture may be needed. In
150
R.M. Smith and F.T. Sheehan
the current state of human kinematic studies, researchers have an array of tools to select from depending on the type of movement they wish to quantify; some techniques only provide static information, while others offer full 3D kinematic data with varying accuracies. To date, CPC MRI is the most accurate means to noninvasively quantify in vivo, 3D musculoskeletal kinematics (Table 3). MRI is the only current technique that can track 3D muscle and skeletal kinematics and does not expose subjects to ionizing radiation. Dynamic MRI is limited by the type of motions that can be studied, due to the typically closed-bore environment. Yet, as scanner technology continues to advance, these limitations will likely be eliminated. Biplane videoradiography has greatly advanced the field of skeletal dynamics. Like MR, it offers a noninvasive methodology to evaluate skeletal kinematics with submillimeter accuracies, but it cannot track muscle movement and does expose subjects to ionizing radiation. The accuracy of these systems depends on the type of movement examined (slow walking vs. running), the shape of the bones under evaluation, and the inherent properties of the capture system. Additionally, the capture FOV and frame rate can limit the types of motion being studied. However, an array of custom-built biplane systems and innovative solutions (e.g., biplane systems mounted on motorized tracks) have emerged to mitigate some of the technical challenges. Lastly, dynamic (four-dimensional) CT is an emerging research methodology. While there is literature describing its qualitative properties, more validation studies are needed to assess the scope of its applicability. One ex vivo study (Zhao et al. 2015) showed promising results for achieving submillimeter accuracy. This review of human motion analysis highlights that each imaging modality has its strengths and limits. Although the measurement technologies reviewed within this chapter have shown and continue to show improved accuracy over time, it is the current trend of combining modalities that will be the true driver of novel information in the coming years. This combination of tools theoretically allows for the best aspects of each technique to be exploited, while downplaying the deficits of each technique. In vivo validation studies that closely mimic the true experimental conditions are crucial for the continued advancement of measuring musculoskeletal motion, regardless of the technique used. It is also important for researchers to report accuracies in a standardized form, such as root mean square or mean absolute errors, and not to simply report bias for previously listed reasons. Finally, through an exploration of these various techniques, it is obvious that the accuracy and precision of each modality should be considered and matched with each study’s intended goals. Acknowledgments We thank Judith Welsh for her help and support toward this project. This work was funded by the Intramural Research Program of the National Institutes of Health Clinical Center, Bethesda, MD, USA. This research was also made possible through the NIH Medical Research Scholars Program, a public-private partnership (http://fnih.org).
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
151
References Abernethy B (2013) Historical origins of the academic study of human movement. In: Abernethy B, Kippers V, Stephanie HJ, Pandy MG, McManus AM, Mackinnon L (eds) Biophysical foundations of human movement. Human Kinetics, Champaign, pp 13–23 Anderst W, Zauel R, Bishop J, Demps E, Tashman S (2009) Validation of three-dimensional modelbased tibio-femoral tracking during running. Med Eng Phys 31(1):10–16. https://doi.org/ 10.1016/j.medengphy.2008.03.003 Asakawa DS, Nayak KS, Blemker SS, Delp SL, Pauly JM, Nishimura DG, Gold GE (2003) Realtime imaging of skeletal muscle velocity. J Magn Reson Imaging: JMRI 18(6):734–739. https:// doi.org/10.1002/jmri.10422 Banks SA, Hodge WA (1996) Accurate measurement of three-dimensional knee replacement kinematics using single-plane fluoroscopy. IEEE Trans Biomed Eng 43(6):638–649. https:// doi.org/10.1109/10.495283 Barrance PJ, Williams GN, Novotny JE, Buchanan TS (2005) J Biomech Eng 127(5):829–837 Behnam AJ, Herzka DA, Sheehan FT (2011) Assessing the accuracy and precision of musculoskeletal motion tracking using cine-PC MRI on a 3.0T platform. J Biomech 44(1):193–197. https://doi.org/10.1016/j.jbiomech.2010.08.029 Bey MJ, Zauel R, Brock SK, Tashman S (2006) Validation of a new model-based tracking technique for measuring three-dimensional, in vivo glenohumeral joint kinematics. J Biomech Eng 128(4):604–609. https://doi.org/10.1115/1.2206199 Bey MJ, Kline SK, Tashman S, Zauel R (2008) Accuracy of biplane x-ray imaging combined with model-based tracking for measuring in-vivo patellofemoral joint motion. J Orthop Surg Res 3:38. https://doi.org/10.1186/1749-799x-3-38 Biswas D, Bible JE, Bohan M, Simpson AK, Whang PG, Grauer JN (2009) Radiation exposure from musculoskeletal computerized tomographic scans. J Bone Joint Surg Am 91(8):1882–1889 Boden BP, Breit I, Sheehan FT (2009) Tibiofemoral alignment: contributing factors to noncontact anterior cruciate ligament injury. J Bone Joint Surg Am 91(10):2381–2389. https://doi.org/ 10.2106/JBJS.H.01721 Bohannon RW, Williams Andrews A (2011) Normal walking speed: a descriptive meta-analysis. Physiotherapy 97(3):182–189. https://doi.org/10.1016/j.physio.2010.12.004 Borotikar BS, Sipprell WH 3rd, Wible EE, Sheehan FT (2012) A methodology to accurately quantify patellofemoral cartilage contact kinematics by combining 3D image shape registration and cine-PC MRI velocity data. J Biomech 45(6):1117–1122 Braekken IH, Majida M, Ellstrom-Engh M, Dietz HP, Umek W, Bo K (2008) Test-retest and intraobserver repeatability of two-, three- and four-dimensional perineal ultrasound of pelvic floor muscle anatomy and function. Int Urogynecol J Pelvic Floor Dysfunct 19(2):227–235. https:// doi.org/10.1007/s00192-007-0408-7 Brainerd EL, Baier DB, Gatesy SM, Hedrick TL, Metzger KA, Gilbert SL, Crisco JJ (2010) X-ray reconstruction of moving morphology (XROMM): precision, accuracy and applications in comparative biomechanics research. J Exp Zool A Ecol Genet Physiol 313(5):262–279. https://doi.org/10.1002/jez.589 Brossmann J, Muhle C, Schroder C, Melchert UH, Bull CC, Spielmann RP, Heller M (1993) Patellar tracking patterns during active and passive knee extension: evaluation with motiontriggered cine MR imaging. Radiology 187(1):205–212. https://doi.org/10.1148/radiology.187. 1.8451415 de Bruin PW, Kaptein BL, Stoel BC, Reiber JH, Rozing PM, Valstar ER (2008) Image-based RSA: roentgen stereophotogrammetric analysis based on 2D-3D image registration. J Biomech 41(1):155–164. https://doi.org/10.1016/j.jbiomech.2007.07.002 Buffi JH, Crisco JJ, Murray WM (2013) A method for defining carpometacarpal joint kinematics from three-dimensional rotations of the metacarpal bones captured in vivo using computed tomography. J Biomech 46(12):2104–2108. https://doi.org/10.1016/j.jbiomech.2013.05.019
152
R.M. Smith and F.T. Sheehan
Burnett KR, Davis CL, Read J (1987) Dynamic display of the temporomandibular joint meniscus by using “fast-scan” MR imaging. AJR Am J Roentgenol 149(5):959–962. https://doi.org/10.2214/ ajr.149.5.959 Byrne CA, Lyons GM, Donnelly AE, O’Keeffe DT, Hermens H, Nene A (2005) Rectus femoris surface myoelectric signal cross-talk during static contractions. J Electromyogr Kinesiol 15(6):564–575. https://doi.org/10.1016/j.jelekin.2005.03.002 Cappozzo A, Paul J (1998) Instrumental observation of human movement: historical development. In: Allard P, Cappozzo A, Lundberg A, Vaughan C (eds) Three-dimensional analysis of human locomotion. New York, NY: Wiley, pp 1–25 Cappozzo A, Catani F, Leardini A, Benedetti MG, Della Croce U (1996) Position and orientation in space of bones during movement: experimental artefacts. Clin Biomech 11(2):90–100. https:// doi.org/10.1016/0268-0033(95)00046-1 Cerciello T, Romano M, Bifulco P, Cesarelli M, Allen R (2011) Advanced template matching method for estimation of intervertebral kinematics of lumbar spine. Med Eng Phys 33(10): 1293–1302. https://doi.org/10.1016/j.medengphy.2011.06.009 Cereatti A, Trojaniello D, Croce UD (2015) Accurately measuring human movement using magneto-inertial sensors: techniques and challenges. In: 2015 I.E. international symposium on Inertial Sensors and Systems (ISISS) proceedings, 23–26 March 2015, Hapuna Beach, HI, USA, pp 1–4. https://doi.org/10.1109/ISISS.2015.7102390 Chao EYS (1980) Justification of triaxial goniometer for the measurement of joint rotation. J Biomech 13(12):989–1006. https://doi.org/10.1016/0021-9290(80)90044-5 Chiari L, Della Croce U, Leardini A, Cappozzo A (2005) Human movement analysis using stereophotogrammetry. Part 2: instrumental errors. Gait Posture 21(2):197–211. https://doi. org/10.1016/j.gaitpost.2004.04.004 Crabolu M, Pani D, Raffo L, Cereatti A (2016) Estimation of the center of rotation using wearable magneto-inertial sensors. J Biomech 49(16):3928–3933. https://doi.org/10.1016/j.jbiomech. 2016.11.046 Dal Maso F, Raison M, Lundberg A, Arndt A, Begon M (2014) Coupling between 3D displacements and rotations at the glenohumeral joint during dynamic tasks in healthy participants. Clin Biomech (Bristol, Avon) 29(9):1048–1055. https://doi.org/10.1016/j.clinbiomech.2014.08.006 Dal Maso F, Blache Y, Raison M, Lundberg A, Begon M (2016) Glenohumeral joint kinematics measured by intracortical pins, reflective markers, and computed tomography: a novel technique to assess acromiohumeral distance. J Electromyogr Kinesiol 29:4–11. https://doi.org/10.1016/j. jelekin.2015.07.008 Drace JE, Pelc NJ (1994) Tracking the motion of skeletal muscle with velocity-encoded MR imaging. J Mag Reson Imaging: JMRI 4(6):773–778 Dupuy D, Hangen D, Zachazewski J, Boland A, Palmer W (1997) Kinematic CT of the patellofemoral joint. AJR Am J Roentgenol 169(1):211–215 Eberhart HD, Inman VT (1951) An evaluation of experimental procedures used in a fundamental study of human locomotion. Ann N Y Acad Sci 51(7):1213–1228 Eng CM, Abrams GD, Smallwood LR, Lieber RL, Ward SR (2007) Muscle geometry affects accuracy of forearm volume determination by magnetic resonance imaging (MRI). J Biomech 40(14):3261–3266. https://doi.org/10.1016/j.jbiomech.2007.04.005 Fellows R, Hill N, Gill H, MacIntyre N, Harrison M, Ellis R, Wilson D (2005) Magnetic resonance imaging for in vivo assessment of three-dimensional patellar tracking. J Biomech 38(8): 1643–1652 Fernandez JW, Akbarshahi M, Kim HJ, Pandy MG (2008) Integrating modelling, motion capture and x-ray fluoroscopy to investigate patellofemoral function during dynamic activity. Comput Methods in Biomech Biomed Eng 11(1):41–53. https://doi.org/10.1080/10255840802296814 Finni T, Hodgson JA, Lai AM, Edgerton VR, Sinha S (2006) Muscle synergism during isometric plantarflexion in achilles tendon rupture patients and in normal subjects revealed by velocityencoded cine phase-contrast MRI. Clin Biomech (Bristol, Avon) 21(1):67–74. https://doi.org/ 10.1016/j.clinbiomech.2005.08.007
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
153
Fischer KJ, Manson TT, Pfaeffle HJ, Tomaino MM, Woo SL (2001) A method for measuring joint kinematics designed for accurate registration of kinematic data to models constructed from CT data. J Biomech 34(3):377–383 Fox AM, Kedgley AE, Lalone EA, Johnson JA, Athwal GS, Jenkyn TR (2011) The effect of decreasing computed tomography dosage on radiostereometric analysis (RSA) accuracy at the glenohumeral joint. J Biomech 44(16):2847–2850. https://doi.org/10.1016/j.jbiomech.2011. 08.009 Fregly BJ, Rahman HA, Banks SA (2005) Theoretical accuracy of model-based shape matching for measuring natural knee kinematics with single-plane fluoroscopy. J Biomech Eng 127(4):692–699 Gondim Teixeira PA, Formery AS, Hossu G, Winninger D, Batch T, Gervaise A, Blum A (2017) Evidence-based recommendations for musculoskeletal kinematic 4D-CT studies using wide area-detector scanners: a phantom study with cadaveric correlation. Eur Radiol 27(2):437–446. https://doi.org/10.1007/s00330-016-4362-y Goto A, Leng S, Sugamoto K, Cooney WP, Kakar S, Zhao K (2014) In vivo pilot study evaluating the thumb carpometacarpal joint during circumduction. Clin Orthop Relat Res 472(4): 1106–1113 Guan S, Gray HA, Keynejad F, Pandy MG (2016) Mobile biplane X-ray imaging system for measuring 3D dynamic joint motion during overground gait. IEEE Trans Med Imaging 35(1):326–336. https://doi.org/10.1109/tmi.2015.2473168 Holden JP, Orsini JA, Siegel KL, Kepple TM, Gerber LH, Stanhope SJ (1997) Surface movement errors in shank kinematics and knee kinetics during gait. Gait Posture 5(3):217–227. https://doi. org/10.1016/S0966-6362(96)01088-0 Im HS, Alter KE, Brochard S, Pons C, Sheehan FT (2014) In vivo pediatric shoulder muscle volumes and their relationship to 3D strength. J Biomech 47(11):2730–2737. https://doi.org/ 10.1016/j.jbiomech.2014.04.049 Jan SVS, Salvia P, Hilal I, Sholukha V, Rooze M, Clapworthy G (2002) Registration of 6-DOFs electrogoniometry and CT medical imaging for 3D joint modeling. J Biomech 35(11): 1475–1484 Jensen ER, Morrow DA, Felmlee JP, Odegard GM, Kaufman KR (2015) Error analysis of cine phase contrast MRI velocity measurements used for strain calculation. J Biomech 48(1):95–103. https://doi.org/10.1016/j.jbiomech.2014.10.035 Jia R, Mellon S, Monk P, Murray D, Noble JA (2016) A computer-aided tracking and motion analysis with ultrasound (CAT & MAUS) system for the description of hip joint kinematics. Int J Comput Assist Radiol Surg 11(11):1965–1977. https://doi.org/10.1007/s11548-016-1443-y Kaiser J, Monawer A, Chaudhary R, Johnson KM, Wieben O, Kijowski R, Thelen DG (2016) Accuracy of model-based tracking of knee kinematics and cartilage contact measured by dynamic volumetric MRI. Med Eng Phys 38(10):1131–1135. https://doi.org/10.1016/j. medengphy.2016.06.016 Kalia V, Obray RW, Filice R, Fayad LM, Murphy K, Carrino JA (2009) Functional joint imaging using 256-MDCT: technical feasibility. Am J Roentgenol 192(6):W295–W299 Karlsson D, Lundberg A (1994) Accuracy estimation of kinematic data derived from bone anchored external markers. In: Proceedings of the 3rd international symposium on 3D analysis of human movement, Stockholm, pp 27–30 Kerkhof FD, Brugman E, D'Agostino P, Dourthe B, van Lenthe GH, Stockmans F, Jonkers I, Vereecke EE (2016) Quantifying thumb opposition kinematics using dynamic computed tomography. J Biomech 49(9):1994–1999. https://doi.org/10.1016/j.jbiomech.2016.05.008 Kettelkamp DB, Johnson RJ, Smidt GL, Chao EY, Walker M (1970) An electrogoniometric study of knee motion in normal gait. J Bone Joint Surg Am 52(4):775–790 Lafortune MA (1984) The use of intra-cortical pins to measure the motion of the knee joint during walking. Doctoral dissertation, Pennsylvania State University, University Park Lafortune MA, Cavanagh PR, Sommer HJ 3rd, Kalenak A (1992) Three-dimensional kinematics of the human knee during walking. J Biomech 25(4):347–357
154
R.M. Smith and F.T. Sheehan
Lafortune MA, Cavanagh PR, Sommer HJ 3rd, Kalenak A (1994) Foot inversion-eversion and knee kinematics during walking. J Orthop Res 12(3):412–420. https://doi.org/10.1002/jor. 1100120314 Lee S, Kim YS, Park CS, Kim KG, Lee YH, Gong HS, Lee HJ, Baek GH (2014) CT-based threedimensional kinematic comparison of dart-throwing motion between wrists with malunited distal radius and contralateral normal wrists. Clin Radiol 69(5):462–467. https://doi.org/ 10.1016/j.crad.2013.09.023 Levens AS, Inman VT, Blosser JA (1948) Transverse rotation of the segments of the lower extremity in locomotion. J Bone Joint Surg Am 30a(4):859–872 Lin CC, Lu TW, Shih TF, Tsai TY, Wang TM, Hsu SJ (2013) Intervertebral anticollision constraints improve out-of-plane translation accuracy of a single-plane fluoroscopy-to-CT registration method for measuring spinal motion. Med Phys 40(3):031912. https://doi.org/10.1118/ 1.4792309 Lippert F, Veress S, Takamoto T, Spolek G (1975) Experimental studies on patellar motion using X-ray photogrammetry. In: Proceedings of symposium on close-range photogrammetric systems, Champaign, IL, USA, pp 186–208 Manal K, McClay I, Stanhope S, Richards J, Galinat B (2000) Comparison of surface mounted markers and attachment methods in estimating tibial rotations during walking: an in vivo study. Gait Posture 11(1):38–45 Maniere-Ezvan A, Havet T, Franconi JM, Quemar JC, de Certaines JD (1999) Cinematic study of temporomandibular joint motion using ultra-fast magnetic resonance imaging. Cranio J Craniomandibular Prac 17(4):262–267 Materials ASfTa (2010) Standard practice for use of the terms precision and bias in ASTM test methods, vol E177-08. ASTM International, West Conshohocken Meskers CG, Fraterman H, van der Helm FC, Vermeulen HM, Rozing PM (1999) Calibration of the “flock of birds” electromagnetic tracking device and its application in shoulder motion studies. J Biomech 32(6):629–633 Milne A, Chess D, Johnson J, King G (1996) Accuracy of an electromagnetic tracking device: a study of the optimal operating range and metal interference. J Biomech 29(6):791–793 Moerman KM, Sprengers AM, Simms CK, Lamerichs RM, Stoker J, Nederveen AJ (2012) Validation of continuously tagged MRI for the measurement of dynamic 3D skeletal muscle tissue deformation. Med Phys 39(4):1793–1810. https://doi.org/10.1118/1.3685579 Neptune RR, Hull ML (1995) Accuracy assessment of methods for determining hip movement in seated cycling. J Biomech 28(4):423–437 Niitsu M, Campeau NG, Holsinger-Bampton AE, Riederer SJ, Ehman RL (1992) Tracking motion with tagged rapid gradient-echo magnetization-prepared MR imaging. J Magn Reson Imaging 2(2):155–163 Ohnishi T, Suzuki M, Nawata A, Naomoto S, Iwasaki T, Haneishi H (2010) Three-dimensional motion study of femur, tibia, and patella at the knee joint from bi-plane fluoroscopy and CT images. Radiol Phys Technol 3(2):151–158. https://doi.org/10.1007/s12194-010-0090-1 Pappas GP, Asakawa DS, Delp SL, Zajac FE, Drace JE (2002) Nonuniform shortening in the biceps brachii during elbow flexion. J Appl Physiol (Bethesda, Md: 1985) 92(6):2381–2389. https:// doi.org/10.1152/japplphysiol.00843.2001 Pelc LR, Sayre J, Yun K, Castro LJ, Herfkens RJ, Miller DC, Pelc NJ (1994) Evaluation of myocardial motion tracking with cine-phase contrast magnetic resonance imaging. Investig Radiol 29(12):1038–1042 Pelc NJ, Drangova M, Pelc LR, Zhu Y, Noll DC, Bowman BS, Herfkens RJ (1995) Tracking of cyclic motion with phase-contrast cine MR velocity data. J Magn Reson Imaging 5(3):339–345 Pipe JG, Boes JL, Chenevert TL (1991) Method for measuring three-dimensional motion with tagged MR imaging. Radiology 181(2):591–595. https://doi.org/10.1148/radiology.181.2. 1924810 Pulkovski N, Schenk P, Maffiuletti NA, Mannion AF (2008) Tissue Doppler imaging for detecting onset of muscle activity. Muscle Nerve 37(5):638–649. https://doi.org/10.1002/mus.20996
Cross-Platform Comparison of Imaging Technologies for Measuring. . .
155
Reinschmidt C, van den Bogert AJ, Lundberg A, Nigg BM, Murphy N, Stacoff A, Stano A (1997) Tibiofemoral and tibiocalcaneal motion during walking: external vs. skeletal markers. Gait Posture 6(2):98–109. https://doi.org/10.1016/S0966-6362(97)01110-7 Rogers B, Wiese S, Blankenbaker D, Meyerand E, Haughton V (2005) Accuracy of an automated method to measure rotations of vertebrae from computerized tomography data. Spine 30(6): 694–696 Salarian A, Horak FB, Zampieri C, Carlson-Kuhta P, Nutt JG, Aminian K (2010) iTUG, a sensitive and reliable measure of mobility. IEEE Trans Neural Syst Rehabil Eng 18(3):303–310. https:// doi.org/10.1109/tnsre.2010.2047606 San Juan JG, Karduna AR (2010) Measuring humeral head translation using fluoroscopy: a validation study. J Biomech 43(4):771–774. https://doi.org/10.1016/j.jbiomech.2009.10.034 Schutzer SF, Ramsby GR, Fulkerson JP (1986) The evaluation of patellofemoral pain using computerized tomography: a preliminary study. Clin Orthop Relat Res 204:288–293 Sheehan FT, Zajac FE, Drace JE (1998) Using cine phase contrast magnetic resonance imaging to non-invasively study in vivo knee dynamics. J Biomech 31(1):21–26 Shih YF, Bull AM, McGregor AH, Humphries K, Amis AA (2003) A technique for the measurement of patellar tracking during weight-bearing activities using ultrasound. Proc Inst Mech Eng H J Eng Med 217(6):449–457 Shih YF, Bull AM, McGregor AH, Amis AA (2004) Active patellar tracking measurement: a novel device using ultrasound. Am J Sports Med 32(5):1209–1217. https://doi.org/10.1177/ 0363546503262693 Simoes MA (2011) Feasibility of wearable sensors to determine Gait parameters. Masters thesis, University of South Florida Sprengers AM, Caan MW, Moerman KM, Nederveen AJ, Lamerichs RM, Stoker J (2013) A scale space based algorithm for automated segmentation of single shot tagged MRI of shearing deformation. Magma (New York, NY) 26(2):229–238. https://doi.org/10.1007/ s10334-012-0332-9 Tang TS, MacIntyre NJ, Gill HS, Fellows RA, Hill NA, Wilson DR, Ellis RE (2004) Accurate assessment of patellar tracking using fiducial and intensity-based fluoroscopic techniques. Med Image Anal 8(3):343–351. https://doi.org/10.1016/j.media.2004.06.011 Tashman S (2008) Comments on “validation of a non-invasive fluoroscopic imaging technique for the measurement of dynamic knee joint motion”. J Biomech 41(15):3290–3291. https://doi.org/ 10.1016/j.jbiomech.2008.07.038. author reply 3292–3293 Tashman S, Anderst W (2003) In-vivo measurement of dynamic joint motion using high speed biplane radiography and CT: application to canine ACL deficiency. J Biomech Eng 125(2): 238–245 Tat J, Kociolek AM, Keir PJ (2015) Validation of color Doppler sonography for evaluating relative displacement between the flexor tendon and subsynovial connective tissue. J Ultrasound Med 34(4):679–687. https://doi.org/10.7863/ultra.34.4.679 Telfer S, Woodburn J, Turner DE (2014) An ultrasound based non-invasive method for the measurement of intrinsic foot kinematics during gait. J Biomech 47(5):1225–1228. https://doi. org/10.1016/j.jbiomech.2013.12.014 Thorhauer E, Tashman S (2015) Validation of a method for combining biplanar radiography and magnetic resonance imaging to estimate knee cartilage contact. Med Eng Phys 37(10):937–947. https://doi.org/10.1016/j.medengphy.2015.07.002 Todorov E (2007) Probabilistic inference of multijoint movements, skeletal parameters and marker attachments from diverse motion capture data. IEEE Trans Biomed Eng 54(11):1927–1939. https://doi.org/10.1109/tbme.2007.903521 Townsend MA, Izak M, Jackson RW (1977) Total motion knee goniometry. J Biomech 10(3):183–193 Tranberg R, Saari T, Zügner R, Kärrholm J (2011) Simultaneous measurements of knee motion using an optical tracking system and radiostereometric analysis (RSA). Acta Orthop 82(2): 171–176. https://doi.org/10.3109/17453674.2011.570675
156
R.M. Smith and F.T. Sheehan
Veress SA, Lippert FG, Hou MC, Takamoto T (1979) Patellar tracking patterns measurement by analytical x-ray photogrammetry. J Biomech 12(9):639–650 Wan EA, Nelson AT (2001) Chapter 5- dual extended Kalman filter methods. In: Haykin S (ed) Kalman filtering and neural networks. New York, NY: Wiley, pp 123–173 Wang S, Passias P, Li G, Li G, Wood K (2008) Measurement of vertebral kinematics using noninvasive image matching method-validation and application. Spine 33(11):E355–E361. https://doi.org/10.1097/BRS.0b013e3181715295 Wang B, Roach KE, Kapron AL, Fiorentino NM, Saltzman CL, Singer M, Anderson AE (2015) Accuracy and feasibility of high-speed dual fluoroscopy and model-based tracking to measure in vivo ankle arthrokinematics. Gait Posture 41(4):888–893. https://doi.org/10.1016/j. gaitpost.2015.03.008 Ward SR, Shellock FG, Terk MR, Salsich GB, Powers CM (2002) Assessment of patellofemoral relationships using kinematic MRI: comparison between qualitative and quantitative methods. J Magn Reson Imaging 16(1):69–74. https://doi.org/10.1002/jmri.10124 Williams AA, Elias JJ, Tanaka MJ, Thawait GK, Demehri S, Carrino JA, Cosgarea AJ (2016) The relationship between tibial tuberosity–trochlear groove distance and abnormal patellar tracking in patients with unilateral patellar instability. Arthrosc J Arthrosc Relat Sur 32(1):55–61 Wilson NA, Press JM, Koh JL, Hendrix RW, Zhang LQ (2009) In vivo noninvasive evaluation of abnormal patellar tracking during squatting in patients with patellofemoral pain. J Bone Joint Surg Am 91(3):558–566. https://doi.org/10.2106/jbjs.g.00572 You BM, Siy P, Anderst W, Tashman S (2001) In vivo measurement of 3-D skeletal kinematics from sequences of biplane radiographs: application to knee kinematics. IEEE Trans Med Imaging 20(6):514–525. https://doi.org/10.1109/42.929617 Zhao K, Breighner R, Holmes D 3rd, Leng S, McCollough C, An KN (2015) A technique for quantifying wrist motion using four-dimensional computed tomography: approach and validation. J Biomech Eng 137(7). https://doi.org/10.1115/1.4030405 Zuhlke T, Fine J, Haughton VM, Anderson PA (2009) Accuracy of dynamic computed tomography to calculate rotation occurring at lumbar spinal motion segments. Spine 34(6):E215–E218. https://doi.org/10.1097/BRS.0b013e318199700d
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and Ligament Glen Lichtwark
Abstract
Ultrasound imaging provides a means to look inside the body and examine how tissues respond to mechanical stress or muscle contraction. As such, it can provide a valuable tool for understanding how muscle, tendon, and ligament mechanics influence the way we move, or vice versa, in health and disease, or to understand how and why these tissues might get injured due to chronic or acute loading. This chapter explores the basic concepts of ultrasound and how it can be used to examine muscle, tendon, and ligament structure and mechanical function. It introduces different techniques, like conventional B-mode imaging, threedimensional ultrasound, and various forms of elastography that can be used to quantify geometrical and mechanical properties of the muscle, tendon, and ligament. Furthermore, methods to quantify muscle and tendon mechanical function during dynamic human movement are explored, and recommendations provided on which techniques are most suitable for different biomechanical investigations. Finally, some predictions about how new ultrasound imaging technologies might continue to advance our understanding of human motion are proposed and explored. Keywords
Biomechanical imaging • Stress • Strain • 3D ultrasound • Tissue tracking • Elastography
G. Lichtwark (*) Centre for Sensorimotor Performance, School of Human Movement and Nutrition Sciences, The University of Queensland, St Lucia, QLD, Australia e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_156
157
158
G. Lichtwark
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Static Measurement of Muscle, Tendon or Ligament Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Basics of Conventional B-Mode Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Measuring Relevant Biomechanical Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Extended Field of View Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three-Dimensional Ultrasound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In Vivo Mechanical Properties of the Muscle, Tendon, and Ligament . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Tissue Response to Forces and/or Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Elastography and Tissue Strain Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shear-Wave Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In Vivo Determination of the Muscle and Tendon Length Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Imaging of Muscle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Imaging of the Tendon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dynamic Imaging of Ligaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other Applications for Ultrasound in Human Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
158 159 159 159 162 163 163 165 165 166 166 168 168 170 172 172 172 173
Introduction A key area of basic, clinical, and applied biomechanics is the quantification of soft tissue mechanical properties and deformation of such tissues in response to force or motion (see chapter ▶ “Simulation of Soft Tissue Loading from Observed Movement Dynamics”). Of particular interest are the mechanical properties of the muscle, tendon, and ligament, as these tissues play a critical role in enabling efficient, powerful, or highly precise movement. In addition, these tissues are also often injured in response to either acute or chronic loads. Ultrasound imaging provides a noninvasive and in vivo method for examining the biomechanical properties and function of the muscle, tendon, and ligament in humans. The dynamic function of the muscle, tendons, and ligament in human motion has traditionally been inferred based on the mechanical properties of tissue that is harvested from human or animal specimens and the estimated forces that these tissues experience during movement. For instance, in a classic study Alexander and Bennet-Clark (1977) estimated the elastic energy that is stored and returned from the Achilles tendon during running by estimating the forces applied to the tendon (through inverse dynamics and estimates of tendon moment arm) and stressstrain properties of the tendon. This approach reveals that much of negative work in early stance is actually stored in the tendon (rather than dissipated by the muscle) and then returned in late stance (rather than being generated by the muscle). Therefore understanding the mechanical interaction of the muscle, tendons, and ligaments is essential to understanding the energetics of human movement. In general, muscles, ligaments, and tendons are commonly characterized by similar compositions across different people (e.g., fiber-type composition in muscle, collagen type and content in the tendon and ligament). However, different
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
159
individuals do display variations in the precise material compositions and structural makeup of the muscle, tendon, and ligament, which can ultimately influence the mechanical function of these tissues. Variation in mechanical properties of soft tissues in the human body are likely to depend on extrinsic factors, like loading commonly experienced by tissues during activities of daily living or athletic training, intrinsic factors like gender, age, body size/structure, and various clinical pathologies or genetic factors. Techniques that can help determine which of these extrinsic and intrinsic factors are important for performance or injury risk is a key area of research in human motion. Ultrasound imaging provides a relatively affordable and low-risk method to look inside the body and examine the mechanical properties of soft tissues like the muscle, tendon, and ligament.
State of the Art Ultrasound imaging can be used for three broad purposes in investigating the mechanics of the muscle, tendon, and ligaments: (1) measurement of tissue architecture to infer mechanical properties, (2) direct (in vivo) determination of tissue mechanical properties, and (3) direct measurement of tissue function during force production or movement. Advances in ultrasound technology, including improvements in image collection and processing, extended field of view measurements, 3D ultrasound imaging, elastography, shear-wave imaging, and advanced methods to track tissue deformation will be discussed below with respect to achieving the above broad purposes.
Static Measurement of Muscle, Tendon or Ligament Architecture A primary determinant of the basic mechanical properties of the muscle, tendon, and ligament is the architecture or geometry of the tissue. For instance, the maximum force generating capacity of muscle is strongly related to its physiological crosssectional area (PCSA, muscle volume divided fiber length (Powell et al. 1984)) and the peak stress of the tendon and ligaments is directly related to its cross-sectional area (Pollock and Shadwick 1994). Ultrasound imaging allows for low-risk, in vivo measurements of the geometry of such tissue, which can be useful for understanding potential performance of such tissues.
The Basics of Conventional B-Mode Ultrasound Ultrasound imaging has been used as a medical diagnosis tool for characterizing tissue dimensions since the 1940s and was quickly adopted for use in musculoskeletal assessment (Kane et al. 2004). Ikai and Fukunaga (1968) used ultrasound to report the relationship between human biceps brachii muscle cross-sectional area
160
G. Lichtwark
Transverse plane (top view)
Sagittal plane (side view)
Image plane
Transducer
Transducer
Fig. 1 Conceptual diagram of scanning plane viewed from the sagittal plane (left) and transverse plane (right), where the transducer is held horizontally to produce a transverse plane image of the underlying tissues (grayscale image overlaid on transverse view). Note that depth is not sufficient to view through the leg and the width is constrained by the size of the transducer
and strength in 1968. However, the systematic use of ultrasound to study the geometry of muscles, tendon, or ligaments for the purpose of understanding human motion and performance really only became prominent in the 1990s (Kallinen and Suominen 1994; Kawakami et al. 1993). Brightness mode, or B-mode, imaging is the most common ultrasound imaging modality, as it can generate an image which is equivalent to a cross section through the tissues in which the ultrasound transducer is imaging (Fig. 1a, b). Tissues that have higher echogenicity (the ability to reflect a sound wave), such as tendinous connective tissue, show up as white on B-mode scans, while tissues with low echogenicity show up as black or gray. As such, contrasts between tissues of different echogenicity are easily identifiable. Muscle fibers typically have low echogenicity, due to their high water content, while the connective tissue that binds the muscle (fascia/aponeurosis) and binds the fibers as fascicles (perimyosim) has high echogenicity (Fig. 2a, b). As such, muscle fascicles can be seen as striated patterns within muscles when the fascicles are viewed within the plane of the image (Fig. 2b). Tendons and ligaments typically have high collagen content, which is relatively highly echogenic; therefore this tissue, and particularly the borders, typically shows up as white pixels in the image (Fig. 2c, d). Ligaments are often less echogenic than tendons, partially because of the less regular filament pattern and the difficulty in aligning to these structures through an image plane. The echogenicity is also dependent on the frequency of the sound waves emitted by the transducer, with superficial structures like the tendon and ligaments optimally imaged with frequencies between 10–12 MHz and deeper muscle at lower frequencies (e.g., 6–8 Hz). The reflection of sound waves depends not only on the composition of the tissue but also on the orientation of the tissue relative to the direction of travel of the sound waves. The “angle of incidence” is the angle at which the sound waves encounter the
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
161
c Tendon – transverse plane
a Muscle – transverse plane Skin
Stand-off pad
Skin/tissue
Subcutaneous fat Muscle tissue
Muscle fascia
Achilles tendon
Muscle tissue
d Tendon – sagittal plane
b Muscle – sagittal plane Tendon Muscle
Achilles tendon Muscle – tendon junction
Muscle tissue
Muscle fascia aponeurosis
Fig. 2 Ultrasound images of the muscle (a–b) and tendon (c–d) in both transverse and sagittal plane relative to the leg. Bright white regions indicate tissues with high echogenicity (e.g., muscle fascia in (a), muscle fascicles in (b)). To ensure contact with the skin with a flat transducer, a standoff pad can be used between the transducer and the tissue, which deforms around the region and allows sound waves to be transmitted, as used in (c)
surface of the tissue of interest (Ihnatsenka and Boezaart 2010) and is optimum when the direction of the structure is approximately perpendicular to the direction of travel of the sound waves. As the angle of incidence increases, so that the sound waves become more and more parallel to the surface of the tissue, the amount of reflection decreases and therefore reduces the definition of the structure within the tissue, as more of the sound waves are scattered and less reflected (Fig. 3a). Therefore to enhance the clarity of structures within an image, the transducer should be positioned such that it is as perpendicular to the structures of interest as possible, or additionally in many modern ultrasound machines the angle at which the sound waves are transmitted can be changed relative to the transducer to enhance the image quality (Fig. 3b). Understanding how the angle of incidence influences image quality is important, because in some instances the surfaces of tissues may have a circular shape (e.g., imaging a tendon cross section, Fig. 3a) or the tissues may move (e.g., during muscle contraction) and hence it is unavoidable that the image quality may not be optimum across the image or over time. Therefore the correct placement of the transducer to maximize the image quality for the purposes of the measurement is essential for measuring muscle, tendon, or ligament architecture or function. Because ligaments often lie between bones, it is often only possible to image superficial ligaments
162
G. Lichtwark
a
b
Transverse view of Achilles tendon (cross section)
Clear echo
Dispersed echo
Conventional
Steering Angle = -10 °
Steering Angle = 10 °
Fig. 3 (a) Conceptual diagram of effect of the angle of incidence or angle of reflecting tissue relative to the sound wave transmission direction. When the tissues are less orthogonal to the sound waves, the reflected sound is dispersed and is not detected by the ultrasound transducers receiving crystals. This can decrease the clarity of borders of tissues that effectively increase the angle of incidence, like tendons in cross section. (b) Pennate muscle fascicles act at an angle to the direction of the sound waves, which reduces the signal strength. Changing the angle of the sound waves through changing the steering angle or by adjusting the direction of the transducer can improve or degrade the clarity of the fascicles, depending on the direction of the fascicles
(e.g., collateral ligaments of the knee) or small part of ligaments between bones. Therefore a critical limiting factor in using ultrasound is its dependence on a good imaging site of the structures relative to the skin, which limits the structures that can be accurately imaged.
Measuring Relevant Biomechanical Parameters The main architectural parameters of interest that can be measured statically in the muscle, tendons, and ligaments are measures of thickness, cross-sectional area, and length (e.g., fascicle or ligament length) (see chapter ▶ “Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion”). The borders of a muscle, tendon, and ligament contain connective tissue that is reasonably echogenic and therefore relatively easily identifiable. Therefore it is often possible to visualize clear cross sections through these structures in different planes (Fig. 2). However, the
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
163
size of the cross section that is visible is limited by the length of the imaging transducer, which defines the maximum width of the image. Therefore for larger muscles or longer tendons (e.g., Achilles) or ligaments (e.g., plantar fascia), it is hard to capture the entire cross section in either the sagittal or transverse planes. While the field of view of ultrasound can limit the potential structures that can be measured, new technologies have provided solutions to overcome this limitation.
Extended Field of View Ultrasound One method to overcome the limitation of the field of view of the transducer is to use ‘extended field of view’ or ‘panoramic’ imaging methods that require the user to move the transducer along a straight line to image consecutive regions of the muscle, which can be stitched back together using image processing techniques (Cooperberg et al. 2001). Such imaging methods have been shown to have good validity and reliability for measuring lengths of muscle fibers and cross-sectional area in various large human muscles and in different planes of imaging (Noorkoiv et al. 2010a, b); however the valid reconstruction of the plane is highly dependent on the ability to move the transducer in a single plane on the surface of the structure.
Three-Dimensional Ultrasound Three-dimensional (3D) imaging of soft tissues makes it possible to make geometrical measurements in 3D space without the limitations of measurements made in a single plane. Freehand 3D ultrasound is a technique that uses conventional two-dimensional imaging (e.g., B-mode), however utilizes multiple image slices along a structure to reconstruct the area of interest, much in the same way that magnetic resonance imaging works. To overcome the issue of the image planes not being perpendicular from one image to the next, it is necessary to track the orientation and position of the transducer and apply a known calibration of the position of the image relative to the transducer, so that images can be correctly projected into the 3D space and accurate voxel information generated (Treece et al. 2003). This method is particularly useful as it allows small spaces between image slices and hence has good spatial resolution, although this is subject to the resolution of the instruments used to track the position and orientation of the transducer (e.g., magnetic or optical systems). Freehand 3D ultrasound also makes it possible to make volumetric measures (e.g., muscle volume; Fig. 4) as well as examine the geometry of large tissue structures that may not be imaged within the field of view of the transducer in conventional B-mode ultrasound [e.g., muscle aponeurosis (Raiteri et al. 2016), tendon length, and cross-sectional area (Obst et al. 2014)]. Three-dimensional ultrasound imaging of smaller volumes can also be performed in real time using specialized transducers. The first method involves capturing a small volume by mechanically sweeping a linear array through an angular motion and
164
G. Lichtwark
b 3D image location
a Transvers scan image
Sagittal reconstruction plane
Transverse image scan
Tibialis anterior muscle border Reconstructed muscle volume
c
Sagittal reconstruction Current image plane
Tibialis anterior muscle border
Fig. 4 Freehand 3D ultrasound uses conventional B-mode images that are collected sequentially along the length of the tissue of interest while the position and orientation of the image is recorded so that the images can be stacked together to generate a 3D volume. (a) Transverse scan of the tibialis anterior muscle in mid-region of muscle. (b) Position of transverse scan relative to muscle and a 3D reconstruction of the muscle volume created from segmenting the muscle borders in sequential images along the muscle. (c) Sagittal plane reconstruction of the muscle through the mid-region
reconstructing the volume in a similar manner to freehand 3D ultrasound. This method is typically used in obstetrics and gynecology, but has found limited use at the present for biomechanics, possibly because of the limited capture volume. However, for imaging small ligamentous structures or sites of muscle injury, this method may be useful because it can collect and reconstruct data in near real time. The second method developed was real-time 3D ultrasound, which uses a matrix of ultrasound elements (instead of an array) to reconstruct volumes at discrete time points; hence this method is often known as four-dimensional (4D) ultrasound (three spatial dimensions and a time dimension). Because of the large amount of data that must be transmitted at high frequencies, the size of the matrix is currently limited and this technology is primarily used to examine cardiac function (e.g., valve mechanics); however there has been some investigation on the function of the pelvic floor muscles (Braekken et al. 2009). This technique should be considered further in both static and dynamic measurement of small muscle, tendon, and ligaments where it is well suited.
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
165
In Vivo Mechanical Properties of the Muscle, Tendon, and Ligament While static measures of soft tissue geometry are valuable for biomechanical assessment of musculoskeletal capacity, the true value of ultrasound comes from being able to determine subject-specific material properties of tissues (see chapters ▶ “Induced Acceleration and Power Analyses of Human Motion,” ▶ “Optimal Control Modeling of Human Movement,” and ▶ “Physics-Based Models for Human Gait Analysis”). Such imaging allows for the characterization of how soft tissues adapt in the event of different loading or exercise, various clinical conditions, and also across the life span. This information is important for understanding capacity to perform movements, requirements for rehabilitative interventions, or prevention of injury.
Dynamic Tissue Response to Forces and/or Movement B-mode ultrasound is the most accessible tool to assess deformation or changes in geometry of human muscle, tendon, and ligaments. Conventional ultrasound machines have good temporal (time) resolution, with frame rates that vary from 10–100 frames per second, depending on imaging parameters, scan depth, and computer processing power. This allows users to record changes in geometry during tasks where tissues are deformed (e.g., muscle contraction, passive stretching). One of the first demonstrations of this capacity was measurement of muscle fascicle length changes during passive length changes (Herbert and Gandevia 1995; Narici et al. 1996) and during isometric contractions (Fukunaga et al. 1997; Fukashiro et al. 1995). These studies clearly showed that human fascicle length changes are not necessarily concomitant with the whole muscle-tendon unit length. This is because the muscle fibers connect to the skeleton via elastic tendons which stretch when force (either passive or active) is applied by the muscle. Utilizing this knowledge, the first in vivo estimates of tendon stiffness in muscles of the lower limb (e.g., gastrocnemius, tibialis anterior) were reported through measuring the shortening of muscle or movement of the muscle-tendon junction during isometric contractions, which was assumed to be the equivalent of the stretch of the elastic tendons (Fukashiro et al. 1995; Maganaris and Paul 1999). Providing that adequate estimates of forces applied to tissues can be determined, typically through measurement of external forces/torques and estimates or direct measurement of muscle moment arms (Maganaris 2005), then these measurements can be used to provide estimates of stress versus strain relationships of both the muscle and tendon. There are numerous limitations to using fascicle length changes during isometric contractions to infer strain or material properties of external tendons. One of the main limitations is that it is difficult to prevent rotation of joints during isometric contractions, which induce fascicle shortening independent of tendon strain (Maganaris 2005; Karamanidis et al. 2005). Another difficulty is the ability to accurately estimate muscle forces using external force measurement techniques
166
G. Lichtwark
(Lichtwark et al. 2013) or ability to accurately synchronize the force and length change data (Finni et al. 2013). Detailed methods to correct for joint rotation errors have been developed (Karamanidis et al. 2005), along with methods to track the movement of the ultrasound transducer to limit errors in displacement measurements and more accurately synchronize signals; however it remains uncertain whether strain measured at one end of a tissue is representative of strain occurring throughout the tissue. In some tendons, like the patella tendon, it is possible to image the entire tendon within the field of the tendon, and hence this has served as a good model to understand tendon adaptation to exercise (Hansen et al. 2006; Onambele et al. 2007; Pearson et al. 2007). However, recent advancements in ultrasound technology have concentrated on methods to examine local strains or material properties of the muscle, tendon, and ligament.
Elastography and Tissue Strain Measurement Ultrasound elastography was first developed as a method to distinguish strain of soft tissues in response to compression. Tissues that strain more for a given compressive force are considered more compliant or elastic. Ultrasound is an ideal medium for examining this mechanical response because the radio frequency (RF) data received from the return sound wave is in the same line of action as the line of compression of the tissue and hence will also compress in response to the tissue deformation. Changes in the compression across a region are indicative of different tissue stiffness values and can be quantified visually using a map overlaid across the image. Methods to compress the tissue are varied and include quasi-static or hand-driven compression, as well as mechanical compression or vibrations or an acoustic radiation force (force generated using the transducer itself) (for reviews of these methods see Varghese 2009; Treece et al. 2011; Nightingale 2011). While all of these methods can be used to quantify the strain of tissues in response to the force, it is difficult to quantify the material properties of the tissue (e.g., Young’s modulus), and these methods only quantify the stiffness of the tissue in the direction of the sound-wave beams. For tissues such as muscles and ligaments, where the interesting material properties are often orthogonal to the sound-wave beams (or parallel with the skin where the line of action of these tissues is usually most prominent), various other exciting methods have been developed.
Shear-Wave Imaging Shear-wave imaging is a form of elastography that utilizes a mechanical perturbation to the tissue, but instead of determining the compression of the tissue, the speed of the resulting propagation of the mechanical shear wave along the tissue is examined through speckle tracking (effective movement of the tissues) across the image. The technique most commonly used to examine the muscle, ligament, and tendon is known as “super-sonic shear imaging” or SSI (Hug et al. 2015). This technique uses
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
167
Fig. 5 Example of supersonic shear imaging (SSI) technique. The bottom image is a conventional B-mode image of the muscle, and the top image has the shear modulus map of a particular region of interest. In this case the muscle is contracting slightly, which increases the shear modulus of the muscle tissue
an acoustic radiation force and ultrafast imaging to quantify the wave propagation speed in real time. Making some assumptions about the tissue density enables a region-specific quantification of shear modulus of the tissue (for full review see Hug et al. (2015), example image in Fig. 5). Quantification of the shear modulus of tissues has been shown to have many biomechanical applications that are useful for understanding human motion. For instance, there is considerable evidence showing that the average shear modulus across an area of muscle is highly related to muscle isometric force across a range of forces that can be assessed (Ates et al. 2015; Bouillard et al. 2012). Therefore, this technique may be a valuable method to understand which muscles contribute to forces that generate human motion (Hug et al. 2015). However, at present the low acquisition rate and low saturation level limit the potential to look at dynamic muscle contractions (Hug et al. 2015). SSI has also been used extensively to examine differences in material properties in the tendon (Helfenstein-Didier et al. 2016;
168
G. Lichtwark
Hug et al. 2013). For instance, the shear modulus of young and old Achilles tendon seems to be different (Slane et al. 2016); however, due to the low saturation level, it is only possible to look at the tendon at very low (often passive) forces. Currently the technology has limitations in being able to understand dynamic muscle, tendon, or ligament function during human motion; however, as this technology improves, it will likely become an invaluable tool for understanding the material properties of the muscle, ligament, and tendon in both healthy and clinical populations.
In Vivo Determination of the Muscle and Tendon Length Changes Ultrasound has a relatively high temporal resolution for characterizing strains of the tissue, and the imaging location is not constrained because the transducer is freely moveable. As such, it is possible to use ultrasound imaging to examine the muscle and tendon length changes during muscular contractions and movement and therefore assess muscle mechanical function of individual muscles.
Dynamic Imaging of Muscle The early measures of muscle fascicle length changes in response to changes in joint angle or isometric contractions clearly demonstrated that the muscle fascicles apparently changed length in a manner that wasn’t consistent with the whole muscletendon unit (Fukashiro et al. 1995; Fukunaga et al. 1997; Kawakami et al. 1998; Narici et al. 1996, Herbert and Gandevia 1995). Since this time, it has become apparent that the length changes of muscle fascicles may be very different to the length changes of the muscle-tendon unit, particularly in the lower limb muscles like the gastrocnemius or soleus. For instance, during the early to mid-stance phase of human walking, it has been demonstrated that muscle fascicles operate relatively isometrically while the muscle-tendon unit is lengthened (Fukunaga et al. 2001; Lichtwark and Wilson 2006). As such, it can be concluded that the tendon tissue must be stretching to store elastic energy that can later be used to help power propulsion. Because muscle fascicles are relatively simple to image through the skin, it is possible to measure dynamic length changes during human motion; however there are numerous technical considerations that must be considered. Firstly, for such imaging to be successful, one must ensure that the ultrasound transducer remains in a similar plane to that in which the fascicles lengthen and shorten. Even in static measures of muscle length, it can be difficult to do this in a way that accurately measures muscle fascicle length. Firstly, this requires that most (if not all) of the length of the muscle fascicle is imaged within the field of view. This limits the size of the muscles that are capable of being imaged. Many of the human lower limb muscles have fiber lengths that are less than the width of the transducer and hence are suitable for dynamic imaging. Secondly, the image plane must be aligned with the line of the fascicles. Recent comparisons between ultrasound imaging in different
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
169
planes and magnetic resonance imaging (more specifically diffusion tensor imaging) in the gastrocnemius muscle concluded that errors in fascicle length measurement of up to 20% could be found with misalignment of the transducer (Bolsterlee et al. 2016a, b). These studies have also found that the best location was approximately perpendicular to the skin, parallel to the tibia, and in the mid-region of the muscle. Whether this alignment remains consistent through a dynamic task is difficult to determine. The general rule of thumb in imaging muscle fascicle lengths during dynamic tasks is that the lines that constitute the connective tissue around the fascicles should remain continuous and clear, as should the fascia to which the muscle connects, throughout the movement (see Fig. 2b). During dynamic contraction the muscle fascicles shorten and their pennation angle increases. As such the image quality decreases due to the change in the angle of incidence (Fig. 3b). This can make it difficult to interpret whether the fascicles are still in the plane of the image. A secure attachment to the site of imaging, such that there is little movement or rotation relative to the skin, is a key requirement. Various flat-shaped ultrasound transducers that can be strapped to the leg seem to be best suited to such tasks (e.g., T-shaped or veterinary rectal transducers). However, as secure attachment requires pressure, this can lead to artificial changes in muscle geometry which need to be considered as they may impact results (Wakeling et al. 2013). Temporal resolution is an important factor when assessing ultrasound data. As is the case with any analysis of human motion, the capture rate must be sufficient to detect the event of interest (typically greater than twice the Nyquist frequency). The rate at which ultrasound machines can collect a complete frame of B-mode ultrasound data depends on many factors including the machine and transducers (processing power, size, and resolution of the transducer) and the imaging settings (e.g., depth of image, image processing techniques). The range of acquisition speeds varies from 5 to 10, 000 frames per second. For slow movements (e.g., passive rotations of joints) a slow frame rate is sufficient, but for activities that require high temporal resolution (e.g., high speed running, impact during landing), a high frame rate is essential. A novel application of ultrahigh speed ultrasound has been the ability to accurately measure the electromechanical delay – the delay between an electrical impulse transmitted along a muscle to induce contraction and the time at which force is transmitted. By examining the timing of length changes of the muscle fibers, which represents force transmission due to stretch of the connective tissue, Nordez and colleagues (Nordez et al. 2009) were able to determine that the electromechanical delay is as small as 6 milliseconds in the ankle plantar flexor muscles, although the time taken to transmit the force along a long tendon like the Achilles may contribute to further delays in force transmission. Automatic tracking of muscle fascicle length changes during dynamic human motion is one major advance that has reduced the time required to assess dynamic length changes. There have been numerous different approaches. The most common approach has been to attempt to track homologous structures from one image frame to the next, either through cross-correlation or optic flow techniques (Korstanje et al.
170
G. Lichtwark
2010; Lee et al. 2008; Loram et al. 2006). These techniques are useful especially for slower movements or those where it is possible to visualize the same structures across consecutive frames. Another method that is commonly used is an optic flow algorithm with an affine fit to estimate deformation across a region of the muscle (Cronin et al. 2011; Farris and Lichtwark 2016; Gillett et al. 2013). Optic flow algorithms create a vector field estimate of the displacement of multiple regions across an image, from one frame to the next. An affine transformation (horizontal and vertical displacement, rotation, dilation, shear in horizontal and vertical direction) can then be fit to the vector field so as to create a smoothed distortion map from one image to the next based on movement across the entire image or area of interest. The distortion map can then be applied to any points within or outside the image, for example, the end points of fascicles, fascia, or other regions of interest. This has been shown to be a useful method for tracking dynamic tasks like walking or isometric contraction. However, the major limitation to these frame-by-frame approaches are that small errors in frame-by-frame estimations of movement can accumulate over time and cause the lengths to drift. High frame rates (such that the movement between frames is small, as is the potential error) can alleviate tracking errors, while other methods have been proposed to correct drift (Farris and Lichtwark 2016); however there are cases where it may be more suitable to treat individual frames as separate problems and to identify structural measurements directly from individual images (e.g., aponeurosis locations, average pennation angle). For instance, there have been attempts to quantify pennation angle automatically in individual frames (Rana et al. 2009) or machine learning algorithms that can detect the length of regions of fascicles (Darby et al. 2012). However such approaches are still subject to noise and can be highly time consuming (e.g., having to train a data set for subsequent analysis) and hence the algorithm of choice is very much dependent on the question being asked and accuracy required.
Dynamic Imaging of the Tendon There has been less focus on examining dynamic function of the tendon when compared to the muscle. As was the case described earlier, much of the literature examining dynamic function of the tendon (e.g., strain responses during movement) have been deduced based on measures of muscle fascicle length changes. Estimates of muscle-tendon length changes are determined using kinematics and various models based on cadaveric data (Hawkins and Hull 1990; Grieve et al. 1978) or geometrical models (Delp et al. 2007), and the difference between length changes of the muscle-tendon unit and the fascicles (often corrected for by the pennation angle) is attributed to the strain of the tendinous tissues (Fukunaga et al. 2001). While this approach gives a global understanding of the function of tendinous tissue, its application to understanding differences between or within different populations is limited because it is hard to assess where potential differences lie within a tendon (regionally within external tendon or within muscle tendinous tissues like aponeurosis).
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
171
One method to determine the length changes of tendons is to examine the movement of the end points. As previously mentioned, in tendons like the patella tendon, that are reasonably short, it may be possible to image the end points of the tendon to assess overall strain (Hansen et al. 2006), although this author is unaware of any publications where this has been done outside of isometric contractions. An alternative for longer tendons, like the Achilles tendon, is to image the muscle-tendon junction (one end point of the tendon, Fig. 2b) and track the position of this junction within the image while also tracking the position and orientation of the image using conventional motion capture techniques (e.g., motion capture markers mounted on the transducer) (Lichtwark and Wilson 2005). This provides a dynamic measure of where the proximal end of the tendon is located, and this can be combined with information about where the tendon inserts onto the skeleton (e.g., marker placed on a bony landmark) and a length between the two points determined. While this is a useful technique for examining tendon strains during dynamic tasks like walking, running, or hopping, it is still limited in that it ignores factors like curvature and/or spiral twists (Obst et al. 2014) in tendons and does not give an indication of regions of high strain along the tendon. One of the most exciting new developments in tendon research has been the development of new elastography methods that use the natural deformation of the tendon during dynamic tasks (i.e., when forces are applied or removed from the tendon) to estimate strain in local regions. Using speckle tracking algorithms that require the raw radio frequency data and which have been customized and validated for estimating tendinous tissue strain (Chernak Slane and Thelen 2014), it has now become possible to examine dynamic tendon strain in the Achilles tendon during dynamic tasks like eccentric contracts (Slane and Thelen 2014) or walking (Franz et al. 2015) (Fig. 6). The potential for this technology to look at region-specific strains (including regions attaching to the gastrocnemius vs. soleus muscles) in various populations (Franz and Thelen 2015) and in different tendons is likely to provide the next major advance in understanding the relationships between movement and tissue strain. However, as was the case with the dynamic muscle imaging, there needs to be numerous considerations
Fig. 6 Example of dynamic speckle tracking (elastography) of the Achilles tendon during dynamic contractions (eccentric). The transducer is placed over along the length of the tendon (imaging in the sagittal plane), and the displacement of individual nodes is tracked using the radio frequency content from the raw ultrasound signal. Note the difference in regional movement from superficial to deep parts of the Achilles tendon (Figure from Slane and Thelen 2014. Permission from Elsevier (License Number 3976471248270))
172
G. Lichtwark
about the potential accuracy of measures depending on factors like transducer movement and misalignment. Dynamic elastography techniques where tissue strain is induced by movement or muscle contraction also has potential for assessing local muscle and ligament strains; however there is currently little or no research in this area.
Dynamic Imaging of Ligaments There is very little literature using ultrasound to examine the dynamic strain of human ligaments. Ultrasound is routinely used as a diagnostic tool to assess ligamentous damage; however there is a dearth of literature on the mechanical function of ligaments in dynamic activities. This may partially be because many ligaments are difficult to image because they reside in locations with high concentrations of bone, which can cause large distortions and unwanted reflections within the image. It may also be because it is difficult to maintain the position of an ultrasound transducer when attached to a joint. Finally, ligaments often twist and turn around bones and hence it may be difficult to align a transducer well with the ligaments. However, the potential to use methods already established in tendon research, including tracking the end points of the bones (Hansen et al. 2006) or using elastography, provides a large avenue of research for understanding ligament function in human motion.
Other Applications for Ultrasound in Human Motion While ultrasound has traditionally been used as an imaging modality to assess soft tissue, it is also increasingly being used to assess bone geometry and bone motion. For instance, 3D ultrasound techniques are being used to assess the location of bony landmarks or bone surfaces (Jia et al. 2016; Passmore and Sangeux 2016) or to assess the position of the joint center of rotation (Peters et al. 2010). Alternatively, measuring the position and orientation of an ultrasound transducer while imaging bony landmarks within the plane of the image can also be used to determine bony translation and/or rotation beneath the surface of the skin (Telfer et al. 2014). This method could potentially be used as a way to correct for soft tissue movement artifact associated with conventional 3D motion analysis techniques (see chapter ▶ “3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers”). The current limitation to these techniques is the small field of view and the fact that only the surface can be imaged due to the fact that ultrasound does not penetrate the bone.
Future Directions Ultrasound imaging has clearly provided a significant advance in the ability to assess the material properties or mechanical function of human muscle, tendon, and ligaments. However, there are numerous limitations to current approaches that either
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
173
limits the capacity to make accurate biomechanical measurements or to fully understand how muscle, tendon, or ligament mechanical properties influence or adapt to human motion. The future will see continued improvement in image resolution (both space and time), potential to track strain or displacement of tissues and methods to perturb tissues for assessment of mechanical properties. Developments in elastography techniques are likely to have the greatest potential in biomechanics. For instance, increases in the rate at which SSI can be collected and the range over which it can measure the shear modulus will allow tissues to be examined while contracting in more realistic conditions and therefore potentially allow researchers to better assess when and how much individual muscles contribute to various tasks. Examining dynamic and localized strains in tendinous and ligamentous tissues using speckle tracking techniques also has great potential at examining injury mechanics in these tissues. One of the biggest limitations of ultrasound is that it is highly localized and only represents a single plane of the tissue of interest. Advancements in 3D technology will be a major area of advance in ultrasound imaging. For instance, 3D elastography has already been developed (Lindop et al. 2006) and may be used to precisely determine areas of tissue “weakness” or structural difference. 4D ultrasound will continue to be developed with larger acquisition volumes that will enable researchers to examine precise muscle, tendon, or ligament strains across multiple planes with good time resolution. A final possibility is that transducers will also advance so that they may also deform along with the tissues of interest to enable more accurate 3D representations with less influence of pressure on the transducer. Despite the relentless advancement of such technology, it is always necessary to validate the ability of these technologies in actually quantifying the measurement of interest. Such validation is often difficult to do in humans in vivo, and hence this is likely to be time-limiting factor in advancing this area of biomechanics and human motion.
References Alexander RM, Bennet-Clark HC (1977) Storage of elastic strain energy in muscle and other tissues. Nature 265:114–117 Ates F, Hug F, Bouillard K, Jubeau M, Frappart T, Couade M, Bercoff J, Nordez A (2015) Muscle shear elastic modulus is linearly related to muscle torque over the entire range of isometric contraction intensity. J Electromyogr Kinesiol 25:703–708 Bolsterlee B, Gandevia SC, Herbert RD (2016a) Effect of transducer orientation on errors in ultrasound image-based measurements of human medial gastrocnemius muscle fascicle length and Pennation. PLoS One 11:e0157273 Bolsterlee B, Gandevia SC, Herbert RD (2016b) Ultrasound imaging of the human medial gastrocnemius muscle: how to orient the transducer so that muscle fascicles lie in the image plane. J Biomech 49:1002–1008 Bouillard K, Hug F, Guevel A, Nordez A (2012) Shear elastic modulus can be used to estimate an index of individual muscle force during a submaximal isometric fatiguing contraction. J Appl Physiol (1985) 113:1353–1361
174
G. Lichtwark
Braekken IH, Majida M, Engh ME, Bo K (2009) Test-retest reliability of pelvic floor muscle contraction measured by 4D ultrasound. Neurourol Urodyn 28:68–73 Chernak Slane L, Thelen DG (2014) The use of 2D ultrasound elastography for measuring tendon motion and strain. J Biomech 47:750–754 Cooperberg PL, Barberie JJ, Wong T, Fix C (2001) Extended field-of-view ultrasound. Semin Ultrasound CT MR 22:65–77 Cronin NJ, Carty CP, Barrett RS, Lichtwark G (2011) Automatic tracking of medial gastrocnemius fascicle length during human locomotion. J Appl Physiol (1985) 111:1491–1496 Darby J, Hodson-Tole EF, Costen N, Loram ID (2012) Automated regional analysis of B-mode ultrasound images of skeletal muscle movement. J Appl Physiol (1985) 112:313–327 Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, Guendelman E, Thelen DG (2007) OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE Trans Biomed Eng 54:1940–1950 Farris DJ, Lichtwark GA (2016) UltraTrack: software for semi-automated tracking of muscle fascicles in sequences of B-mode ultrasound images. Comput Methods Prog Biomed 128:111–118 Finni T, Peltonen J, Stenroth L, Cronin NJ (2013) Viewpoint: on the hysteresis in the human Achilles tendon. J Appl Physiol (1985) 114:515–517 Franz JR, Slane LC, Rasske K, Thelen DG (2015) Non-uniform in vivo deformations of the human Achilles tendon during walking. Gait Posture 41:192–197 Franz JR, Thelen DG (2015) Depth-dependent variations in Achilles tendon deformations with age are associated with reduced plantarflexor performance during walking. J Appl Physiol (1985) 119:242–249 Fukashiro S, Itoh M, Ichinose Y, Kawakami Y, Fukunaga T (1995) Ultrasonography gives directly but noninvasively elastic characteristic of human tendon in vivo. Eur J Appl Physiol Occup Physiol 71:555–557 Fukunaga T, Ichinose Y, Ito M, Kawakami Y, Fukashiro S (1997) Determination of fascicle length and pennation in a contracting human muscle in vivo. J Appl Physiol (1985) 82:354–358 Fukunaga T, Kubo K, Kawakami Y, Fukashiro S, Kanehisa H, Maganaris CN (2001) In vivo behaviour of human muscle tendon during walking. Proc Biol Sci 268:229–233 Gillett JG, Barrett RS, Lichtwark GA (2013) Reliability and accuracy of an automated tracking algorithm to measure controlled passive and active muscle fascicle length changes from ultrasound. Comput Methods Biomech Biomed Engin 16:678–687 Grieve DW, Pheasant S, Cavanagh PR (1978) Prediction of gastrocnemius length from knee and ankle joint posture. Biomechanics vi-a. University Park Press, Baltimore Hansen P, Bojsen-Moller J, Aagaard P, Kjaer M, Magnusson SP (2006) Mechanical properties of the human patellar tendon, in vivo. Clin Biomech (Bristol, Avon) 21:54–58 Hawkins D, Hull ML (1990) A method for determining lower extremity muscle-tendon lengths during flexion/extension movements. J Biomech 23:487–494 Helfenstein-Didier C, Andrade RJ, Brum J, Hug F, Tanter M, Nordez A, Gennisson JL (2016) In vivo quantification of the shear modulus of the human Achilles tendon during passive loading using shear wave dispersion analysis. Phys Med Biol 61:2485–2496 Herbert RD, Gandevia SC (1995) Changes in pennation with joint angle and muscle torque: in vivo measurements in human brachialis muscle. J Physiol 484(Pt 2):523–532 Hug F, Lacourpaille L, Maisetti O, Nordez A (2013) Slack length of gastrocnemius medialis and Achilles tendon occurs at different ankle angles. J Biomech 46:2534–2538 Hug F, Tucker K, Gennisson JL, Tanter M, Nordez A (2015) Elastography for muscle biomechanics: toward the estimation of individual muscle force. Exerc Sport Sci Rev 43:125–133 Ihnatsenka B, Boezaart AP (2010) Ultrasound: basic understanding and learning the language. Int J Shoulder Surg 4:55–62 Ikai M, Fukunaga T (1968) Calculation of muscle strength per unit cross-sectional area of human muscle by means of ultrasonic measurement. Int Z Angew Physiol 26:26–32
Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and. . .
175
Jia R, Mellon S, Monk P, Murray D, Noble JA (2016) A computer-aided tracking and motion analysis with ultrasound (CAT & MAUS) system for the description of hip joint kinematics. Int J Comput Assist Radiol Surg 11:1965–1977 Kallinen M, Suominen H (1994) Ultrasonographic measurements of the Achilles tendon in elderly athletes and sedentary men. Acta Radiol 35:560–563 Kane D, Grassi W, Sturrock R, Balint PV (2004) A brief history of musculoskeletal ultrasound: ‘from bats and ships to babies and hips’. Rheumatology (Oxford) 43:931–933 Karamanidis K, Stafilidis S, Demonte G, Morey-Klapsing G, Bruggemann GP, Arampatzis A (2005) Inevitable joint angular rotation affects muscle architecture during isometric contraction. J Electromyogr Kinesiol 15:608–616 Kawakami Y, Abe T, Fukunaga T (1993) Muscle-fiber pennation angles are greater in hypertrophied than in normal muscles. J Appl Physiol (1985a) 74:2740–2744 Kawakami Y, Ichinose Y, Fukunaga T (1998) Architectural and functional features of human triceps surae muscles during contraction. J Appl Physiol (1985b) 85:398–404 Korstanje JW, Selles RW, Stam HJ, Hovius SE, Bosch JG (2010) Development and validation of ultrasound speckle tracking to quantify tendon displacement. J Biomech 43:1373–1379 Lee SS, Lewis GS, Piazza SJ (2008) An algorithm for automated analysis of ultrasound images to measure tendon excursion in vivo. J Appl Biomech 24:75–82 Lichtwark GA, Cresswell AG, Ker RF, Reeves ND, Maganaris CN, Magnusson SP, Svensson RB, Coupe C, Hershenhan A, Eliasson P, Nordez A, Foure A, Cornu C, Arampatzis A, MoreyKlapsing G, Mademli L, Karamanidis K, Vagula MC, Nelatury SR (2013) Commentaries on viewpoint: on the hysteresis in the human Achilles tendon. J Appl Physiol (1985) 114:518–520 Lichtwark GA, Wilson AM (2005) In vivo mechanical properties of the human Achilles tendon during one-legged hopping. J Exp Biol 208:4715–4725 Lichtwark GA, Wilson AM (2006) Interactions between the human gastrocnemius muscle and the Achilles tendon during incline, level and decline locomotion. J Exp Biol 209:4379–4388 Lindop JE, Treece GM, Gee AH, Prager RW (2006) 3D elastography using freehand ultrasound. Ultrasound Med Biol 32:529–545 Loram ID, Maganaris CN, Lakie M (2006) Use of ultrasound to make noninvasive in vivo measurement of continuous changes in human muscle contractile length. J Appl Physiol (1985) 100:1311–1323 Maganaris CN (2005) Validity of procedures involved in ultrasound-based measurement of human plantarflexor tendon elongation on contraction. J Biomech 38:9–13 Maganaris CN, Paul JP (1999) In vivo human tendon mechanical properties. J Physiol 521(Pt 1):307–313 Narici MV, Binzoni T, Hiltbrand E, Fasel J, Terrier F, Cerretelli P (1996) In vivo human gastrocnemius architecture with changing joint angle at rest and during graded isometric contraction. J Physiol 496(Pt 1):287–297 Nightingale K (2011) Acoustic radiation force impulse (ARFI) imaging: a review. Curr Med Imaging Rev 7:328–339 Noorkoiv M, Nosaka K, Blazevich AJ (2010a) Assessment of quadriceps muscle cross-sectional area by ultrasound extended-field-of-view imaging. Eur J Appl Physiol 109:631–639 Noorkoiv M, Stavnsbo A, Aagaard P, Blazevich AJ (2010b) In vivo assessment of muscle fascicle length by extended field-of-view ultrasonography. J Appl Physiol (1985) 109:1974–1979 Nordez A, Gallot T, Catheline S, Guevel A, Cornu C, Hug F (2009) Electromechanical delay revisited using very high frame rate ultrasound. J Appl Physiol (1985) 106:1970–1975 Obst SJ, Newsham-West R, Barrett RS (2014) In vivo measurement of human achilles tendon morphology using freehand 3-D ultrasound. Ultrasound Med Biol 40:62–70 Onambele GN, Burgess K, Pearson SJ (2007) Gender-specific in vivo measurement of the structural and mechanical properties of the human patellar tendon. J Orthop Res 25:1635–1642 Passmore E, Sangeux M (2016) Defining the medial-lateral axis of an anatomical femur coordinate system using freehand 3D ultrasound imaging. Gait Posture 45:211–216
176
G. Lichtwark
Pearson SJ, Burgess K, Onambele GN (2007) Creep and the in vivo assessment of human patellar tendon mechanical properties. Clin Biomech (Bristol, Avon) 22:712–717 Peters A, Baker R, Sangeux M (2010) Validation of 3-D freehand ultrasound for the determination of the hip joint centre. Gait Posture 31:530–532 Pollock CM, Shadwick RE (1994) Allometry of muscle, tendon, and elastic energy storage capacity in mammals. Am J Phys 266:R1022–R1031 Powell PL, Roy RR, Kanim P, Bello MA, Edgerton VR (1984) Predictability of skeletal muscle tension from architectural determinations in guinea pig hindlimbs. J Appl Physiol Respir Environ Exerc Physiol 57:1715–1721 Raiteri BJ, Cresswell AG, Lichtwark GA (2016) Three-dimensional geometrical changes of the human tibialis anterior muscle and its central aponeurosis measured with three-dimensional ultrasound during isometric contractions. PeerJ 4:e2260 Rana M, Hamarneh G, Wakeling JM (2009) Automated tracking of muscle fascicle orientation in B-mode ultrasound images. J Biomech 42:2068–2073 Slane LC, Thelen DG (2014) Non-uniform displacements within the Achilles tendon observed during passive and eccentric loading. J Biomech 47:2831–2835 Slane LC, Martin J, Dewall R, Thelen D, Lee K (2016) Quantitative ultrasound mapping of regional variations in shear wave speeds of the aging Achilles tendon. Eur Radiol 27(2):474–482 Telfer S, Woodburn J, Turner DE (2014) An ultrasound based non-invasive method for the measurement of intrinsic foot kinematics during gait. J Biomech 47:1225–1228 Treece GM, Gee AH, Prager RW, Cash CJ, Berman LH (2003) High-definition freehand 3-D ultrasound. Ultrasound Med Biol 29:529–546 Treece G, Lindop J, Chen L, Housden J, Prager R, Gee A (2011) Real-time quasi-static ultrasound elastography. Interface Focus 1:540–552 Varghese T (2009) Quasi-static ultrasound Elastography. Ultrasound Clin 4:323–338 Wakeling JM, Jackman M, Namburete AI (2013) The effect of external compression on the mechanics of muscle contraction. J Appl Biomech 29:360–364
Part III Generative Methods in Dynamic Pose Estimation
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers Thomas M. Kepple and Alan R. De Asha
Abstract
Position and orientation (Pose) estimations of the human body during motion that are derived from data collected using any marker-based camera system have inherent errors related to a combination of measurement noise, soft tissue artifact (STA), and inaccuracies due to incorrect marker placement. Individually, and in combination, these errors reduce the overall accuracy of marker-based Pose estimation. Optimization and multibody dynamics methods have been formulated to reduce these errors. However it has been argued that uncertainty in data, such as that caused by sensor noise, soft tissue deformation, marker movement, or inaccurate marker placement, cannot be directly accounted for using traditional deterministic approaches. We postulate that uncertainty can be more appropriately addressed by casting the Pose estimation problem within the general framework of probabilistic inference. In this chapter, we will introduce Bayes theorem, the basis for probabilistic inference, and give a general example of how a Bayesian approach can take advantage of prior knowledge to improve estimation. We will then formulate Bayes theorem in the context of mitigating uncertain marker motion. Finally, we will apply this approach on some sample data to demonstrate how this method can, in practice, produce substantially better measurement of knee joint motion then the previously established deterministic methods. Keywords
Bayesian inference • Pose estimation • Motion capture • Markers • Multibody models • Probabilistic • X-ray
T.M. Kepple (*) • A.R. De Asha C-Motion Inc., Germantown, MD, USA e-mail: [email protected]; [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_158
179
180
T.M. Kepple and A.R. De Asha
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Probabilistic Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bayes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Probabilistic Pose Estimation from Marker-Based Data: Theory . . . . . . . . . . . . . . . . . . . . . . . . . . The Importance of the Covariance Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Probabilistic Pose Estimation from Marker-Based Data: An Example . . . . . . . . . . . . . . . . . . . . . Casting the 6-DOF and IK Methods in a Statistical/Generative Framework . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
180 181 182 182 184 186 187 192 193 194
Introduction For those of us who have always used discriminative or deterministic models for solving the Pose estimation problem, the generative approach about to be described requires a conceptual leap. The solution to the position and orientation (Pose) of a model, for a given set of data, is, perhaps counterintuitively, not to solve for the Pose directly. Instead it is to solve for the possible data sets that are consistent with the measured Pose. In this chapter, we consider the Pose estimations of multi-segment, rigid body (multibody) models based on recordings of 3D optical marker-based motion data. The principal assumption of these Pose estimation algorithms (▶ “3D Dynamic Pose Estimation from Marker-Based Optical Data”) is that the markers move rigidly along with the body segments to which they are attached. It is known, however, that marker measurements have errors related to noise and marker movement relative to the underlying skeleton (soft tissue artifact, STA) (▶ “Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis”) and can have inaccuracies due to incorrect marker placement (Leardini et al. 2005; Taylor et al. 2005; Peters et al. 2010). All of these errors reduce the accuracy of the Pose estimation. The consequence of STA alone is that the accuracy with which bone motion can be measured by noninvasive clinical motion analysis is typically insufficient for tissue-/joint-level biomechanical analysis (Cappello et al. 2005). It should be noted that these nonrigid marker motions can be represented mathematically (Dumas et al. 2014; Grimpampi et al. 2014), but it has proven difficult to incorporate this information into a deterministic Pose estimation algorithm. It has been argued (Todorov 2007) that uncertainties related to STA cannot be mitigated directly from the currently used discriminative Pose estimation methods and that Pose estimation from motion capture (MoCap) data is best estimated by assuming uncertainty in the data. Todorov suggested using well-established probabilistic algorithms, based on Bayesian inference. These algorithms provide a principled way for making optimal inferences from uncertain data in combination with previous experience. Probabilistic Pose estimation may be understood conceptually by realizing that an improved estimate of the Pose of the model for any given set of
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
181
data is not to solve for the Pose directly (a deterministic or discriminative model) but rather to solve for the Pose using a generative model that minimizes a combination of the estimated errors of a discriminative Pose and a predicted Pose. Probabilistic Pose estimation provides a principled way to include models of STA in the Pose estimation algorithms and to produce better Pose estimates of the skeleton.
State of the Art Current methodologies for estimating the Pose of rigid bodies from marker data, such as those described in the previous chapter, are still based on deterministic methodologies. The simplest deterministic approaches to 3D movement analysis use direct methods (Kadaba et al. 1989; Davis et al. 1991) that are based on vector algebra to compute the Pose of a model, made up of rigid anatomical segments. Spoor and Veldpaus (1980) cast Pose estimation as an optimization problem in order to try and mitigate STA and other measurement errors. Their optimization approach was based on the assumption that the configuration of targets on a segment remains constant with respect to each other and with respect to the underlying bone. Andriacchi et al. (1998) enhanced the segmental optimization (6 DOF) approach by distributing a large number of markers on a given segment and reducing the contribution of unreliable markers to the solution in the hope of further reducing the effects of STA. Lu and O’Connor (1999) extended the segment optimization approach to a multibody optimization, or inverse kinematic (IK), solution discussed in the previous chapter, by adding joint constraints to the model in order to further reduce errors. Other methods, based on dynamics, including residual elimination (Remey and Thelen 2009), residual reduction (Hamner et al. 2008), and optimal control (Kaplan and Heegaard 2001; van den Bogert et al. 2011; Miller et al. 2016; Koelewijin et al. 2016) have also been used to improve estimations of Pose. In particular, optimal control is an emerging method for treating multibody Pose estimation (Kaplan and Heegaard 2001; van den Bogert et al. 2011; Miller and Hamill 2015) (chapter ▶ “Optimal Control Modeling of Human Movement”; ▶ “Physics-Based Models for Human Gait Analysis”). The objective is to determine the controls, u(t), which produce a movement that is, in some sense, “optimal.” Optimal control casts Pose estimation as a trade-off between an initial estimate of the model’s Pose and multibody dynamics. Multibody model dynamics can be represented symbolically by the function f of the model’s state variables, s, and control variables, u (i.e., the equations of motion): f ðs, s0 , uÞ ¼ 0
(1)
where the controls u specify the joint moments and the state vector s that contains the model’s generalized coordinates (the Pose) and generalized speeds. The tracking error in the equations of motion between the model (s) and motion capture (q) generalized coordinates is given by:
182
T.M. Kepple and A.R. De Asha
dðs, uÞ ¼ q
ðT 0
f ðs, s0 , uÞ
(2)
d(s, u) represents the difference between the Pose Ð T estimate generated from the motion capture data q and the simulated Pose, 0 f ðs, s0 , uÞ , estimated from the multibody model at time T. These optimal control-based methods generally require either additional data from one or more force platforms (GRF) or an anatomically congruent full-body model. For example, the equations of motion typically contain a discrete-element viscoelastic/Coulomb friction model of foot-ground contact (Miller and Hamill 2015). An additional error term can then be represented as the difference between the model-generated ground reaction force (GRF) EGRF and the measured GRF (GRF(t)): gðs, uÞ ¼ EGRF ðs, τðuÞÞ GRFðtÞ
(3)
g(s, u) is added as a constraint on the objective function to be minimized in the optimization. The result is that the optimal control solutions do not supply a general solution for tracking rigid bodies using surface markers. Because these dynamics-based methods generally require either additional data from one or more force platforms or an anatomically congruent full-body model, they do not supply a general solution for tracking rigid bodies using surface markers. Probabilistic inference, unlike the deterministic approaches of optimal control or 6 DOF and IK (described in the previous chapter), is able to deal with uncertainty and thereby makes optimal use of available data. The “noise” is captured by a generative model, which defines the conditional probability of the data given the Pose. The most advanced use of probabilistic Pose estimation is a new more generalized approach to Pose estimation based on real-time state estimation (Lowrey et al. 2017). This approach, however, is currently used in robotics not human motion measurement.
Probabilistic Pose Estimation Bayes Theorem Estimating the Pose of a model probabilistically begins with Bayes theorem: Pðqj vÞ ¼
pðvj qÞpðqÞ pð v Þ
(4)
To explain the basis of Bayes theorem, we shall step through a simple example. A female soccer player comes into the office of a sports medicine doctor complaining of knee pain. The doctor conducts an anterior draw test (an imperfect
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
183
diagnostic test) on the patient and finds more laxity in the painful knee than in the unaffected knee. What is the probability that the patient has an ACL injury? Using a Bayesian approach to the problem, we start by applying the Bayesian likelihood: p(v|q), that is, we know that the probability of the data, q (a positive anterior draw test), given the hypothesis, v (an ACL injury), is: pðvj qÞ ¼ 0:55 Using our prior knowledge of ACL injury, p(q), that female soccer players are more likely to suffer ACL injuries than participants in other sports, we use this past experience to determine the probability that our soccer-playing patient has an ACL injury: pðqÞ ¼ 0:65 Finally, we need to find the probability of the data, the normalization term, p(v). The normalization term is based on the probability that the patient is not ACL injured ( p(~q) = 1.0 0.65 = 0.35). We also need the probability that a patient has an ACL injury but is not a female soccer player. (For our example, we assume this value: p(v| q) = 0.5). To compute the normalization term, we use: pðvÞ ¼ pðvj qÞ pðqÞ þ pð qÞ pðvj qÞ
(5)
which for our example is: pðvÞ ¼ ð0:55Þ ð0:65Þ þ ð0:35Þ 0:5 ¼ 0:5325 So Bayes theorem becomes: Pðqj vÞ ¼
0:55 0:65 ¼ 0:67 0:5325
Note that in this case we use our prior knowledge, that a female soccer player presenting with knee pain is more likely to have an ACL injury than another athlete, to improve the predictability of the anterior draw test from 0.55 to 0.67. It is important to note that in order for this approach to work properly, this prior must be veridical, not simply based on personal bias; that is, it should come from some valid source of prior knowledge (published epidemiological research would be a valid prior source in our example). In this way, the prior will improve accuracy while minimizing any reduction in the value of available data to the outcome. Armed with this understanding of how Bayes theorem works, we will now apply these same principles to use prior knowledge about how markers move relative to underlying bone (i.e., STA). We shall do this in order to obtain better estimates of a model’s Pose using data from a motion capture system, by mitigating the effects of STA.
184
T.M. Kepple and A.R. De Asha
Probabilistic Pose Estimation from Marker-Based Data: Theory Throughout this chapter, we discuss the application of Bayes theorem to improve Pose estimation for traditional marker-based MoCap data. To improve Pose estimates from MoCap data (v), the terms in Bayes theorem (Eq. 4) become: P(q| v) = the posterior, which is the probability of the Pose, q, given the marker data, v P(v| q) = the likelihood, which is the probability of the marker data, v, given the Pose, q P(q) = the prior which is the initial probability of the Pose, q, given some prior knowledge of the model (i.e., an expected state) p(v) = the probability of the data or normalization term. In our probabilistic inference approach, the normalization term can be ignored because it is considered to be a constant within an optimization problem (Todorov 2007). To frame our Pose estimation problem via Bayes theorem, we start by defining the likelihood from a discriminative IK Pose estimation. From IK (Lu and O’Connor 1999), each marker has a static vector ai that represents the location of the marker in the anatomical reference frame of a segment (AF) and a dynamic vector vi that represents the location of the marker in the global reference frame (GF) (for markers i = 1,2,. . ..N). The transformation for any marker i from AF to GF is given by a set of rotation matrices R and translation vectors O, which are a function of the model’s total generalized coordinates q: vi ¼ RðqÞai þ OðqÞ
(6)
Due to measurement error and STA, the location of markers relative to the underlying bone will change during movement, thus there will always be an error vector (residual) r ðq, vi Þ for each marker: r ðq, vi Þ ¼ vi RðqÞai þ OðqÞ
(7)
We will assume that all error r ðq, vi Þ comes from marker uncertainty and that the error is normal and independent. For convenience, we define the matrix v which contains all components from all of the marker vectors. For consistency with the probabilistic Pose estimation, the IK residuals are represented as a generative model, with the IK residuals taking the form of the conditional probability distribution: Pðvj qÞ N d ðr ðq, vÞ, V Þ
(8)
where Nd(r(q, v), V ) is a multivariate normal distribution centered around the mean of residual error r(q, v) with a covariance matrix V generated for all components of each marker. Expressing the likelihood as a Gaussian distribution in canonical form (Stroupe et al. 2001)
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
Pðvj qÞ ¼
1 1 pffiffiffiffiffiffi eð2Þ 2Π jV j
rðq, vÞ
T
V 1 rðq, vÞ
185
(9)
To find the best Pose estimate, we define our problem as solving q for the maximum likelihood P(v| q). For convenience, we take the negative log of the likelihood: logðPðvj qÞÞ ¼ C þ
1 ðr ðq, vÞÞT V 1 ðr ðq, vÞÞ 2
(10)
Maximizing P(v| q) is now equivalent to finding q to minimize log (P(v| q), for the set of marker coordinates, v (Todorov 2007), which we express as: f ðqÞ ¼ r ðq, vÞT V 1 r ðq, vÞ
(11)
In other words, the solution to the IK Pose estimation is to find q that minimizes Eq. 11 over all marker data, v. The fundamental advantage of the probabilistic Pose estimation is that we can introduce prior knowledge of biomechanics in a principled way. We can represent this knowledge of the state of the system (^q) as the expected Pose estimate obtained from joint constraint(s) and/or a simple STA model. The difference between the current state of the system and the expected state is: Prior error ¼ ðq ^ qÞ
(12)
Consistent with our expectation of the uncertainty in the Pose estimate, we assume that the differences between our current and predicted state estimates, ðq ^ q Þ are normal and independent with a covariance of U: Pðq ^q Þ N d ðr ðq ^ q Þ, U Þ
(13)
Expressing the prior as a Gaussian distribution in canonical form (Stroupe et al. 2001) Pðq ^q Þ ¼
1 1 pffiffiffiffiffiffi eð2Þ 2Π jV j
rðq^ qÞ
T
U1 rðq^q Þ
(14)
Referring back to Bayes theorem (Eq. 4), if we declare that both our likelihood and prior are Gaussian distributions and express them in canonical form, P(v| q) takes the form: Pðvj qÞ ¼
T 1 1 T 1 1 1 1 pffiffiffiffiffiffi eð2Þ rðq, vÞ V rðv, xÞ pffiffiffiffiffiffiffi eð2Þðq^q Þ U ðq^q Þ 2Π jV j 2Π jU j
(15)
186
T.M. Kepple and A.R. De Asha
where ^ q were the predicted positional state variables, V1 is the inverted covariance matrix obtained from the variation of the IK residuals (Likelihood) over multiple trials of data (i.e., not the identity matrix), and U1 is the inverted covariance matrix of the difference between measured states q and expected states ^q . As before, we solve for the maximum likelihood P(v| q) by minimizing the negative log of: 1 ðr ðq, vÞÞT V 1 ðr ðq, vÞÞ 2 1 þ ðq ^q ÞT U 1 ðq ^q Þ 2
logðPðvj qÞÞ ¼ C þ
(16)
which we express as: f ðqÞ ¼ r ðq, vÞT V 1 r ðq, vÞ þ ðq ^ q ÞT U 1 ðq ^q Þ
(17)
So the solution to the Bayesian Pose estimation is to find q that minimizes Eq. 17 over all marker data, v.
The Importance of the Covariance Matrices In Eq. 17, the covariance matrices, V1 and U1, serve two vital roles. Firstly, consider that r(q, v) is a measure of the marker residual error (Eq. 7), and thus this difference between the measured and expected marker locations will always be a distance. Conversely, ðq ^q Þ is a measure of the difference between the measured and expected values of the Pose. Note the model Pose, q, will contain both position and orientation degrees of freedom, and therefore the values in ðq ^q Þ (Eq. 12) can appear as both distances and angular measures. Thus, since the two terms in Eq. 17 contain matrices whose elements may have different units, these elements need to be normalized to allow the two terms in the cost function to be added. Since V is a measure of the covariance of r(q, v), the elements of variance matrix V will have the same units as r(q, v). Likewise, all the elements of U are obtained from the variance in ðq ^ q Þ and thus have the same units as the elements of ðq ^q Þ. From this, we can see that inverting these matrices, V1 and U1, will normalize all the elements in Eq. 17 so that they can be used in a meaningful cost function. The second role for the covariance matrices comes from the fact that, at each time step, V1 will weigh the solution to Eq. 17 inversely to the variation found by the likelihood r(q, v), while U1 will weigh the solution to Eq. 17 inversely to the variation expected by the prior (q ^q ). Thus if the data for either the likelihood or the prior at any given instant during the movement become more variable, we tend to trust those data less. This means they will contribute less to the overall solution. Conversely, when the variation in the data decreases, the values in the inverted covariance matrices will increase, and these data will get extra weight in the solution.
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
187
Since our assumption is that both our prior and likelihood errors are normally distributed (Eqs. 7 and 12), weighing the data in this manner is consistent with Bayesian inference. Practically speaking, in order to obtain the covariance matrices V1 and U1, it is best to capture a large number of independent trials from a group of subjects that matches the population of interest. The data from these trials can then be time normalized to a fixed number of frames between the start and the end of the movement. The covariance of the fit of the MoCap data to the expected target locations, as obtained from Eq. 7, gives us the matrix V. Likewise, at each time step during the motion, we have both a measured and an expected (prior) value for model Pose, q, with the difference between the two being given by Eq. 12. We obtain the covariance matrix, U, by applying Eq. 12 at each time step over our large sample size of subjects. Finally by inverting both Vand U, we get V1 and U1 which weight the solution to Eq. 17 (under the assumption that our errors are normally distributed).
Probabilistic Pose Estimation from Marker-Based Data: An Example Now that we have considered and discussed this probabilistic theory, let’s investigate some data to illustrate how it can be used in practice. The test data for this chapter were collected in collaboration with Dr. Scott Tashman at the biodynamics laboratory of the University of Pittsburgh. Three subjects walked on a treadmill while data were captured synchronously using a traditional marker-based motion capture system (a synchronized 12 camera Vicon MX system) with a conventional lower extremity gait marker set and a 3D dynamic stereo X-ray (DSX) system (chapter ▶ “Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography”). The DSX system had been previously validated, both in static and dynamic joint rotation conditions (Anderst et al. 2009). For the static testing, the DSX translational accuracy was found to be on the order of 0.2 mm and rotational accuracy on the order of 0.2 degrees. During the dynamic testing, the accuracy of the DSX system was found to be better than 1.0 degree for all rotations (flexion/extension, ab/adduction, and axial rotation), and translational accuracy was found to be better than 0.7 mm in all three planes. Thus the DSX system serves as our “gold standard” throughout this example. A calibration cube (11.5 cm per side), containing both tantalum beads and optical motion capture markers, was used to calibrate the DSX and also to define the transformation between the DSX and motion capture space. Shank and thigh positions and orientations were measured by both the DSX system and markerbased MoCap system. We compared the DSX Pose values against three different marker-based Pose estimation algorithms: 6-DOF, IK, and a probabilistic method. The errors associated with the different marker-based methods were computed from the RMS translation errors, determined by the Euclidean distance between the segment origins obtained from the MoCap data and from the “gold standard” DSX system.
188
T.M. Kepple and A.R. De Asha
As noted above, our probabilistic inference approach relies on having some prior knowledge about the measured motion. The “biomechanically inspired” prior used in this probabilistic Pose example was based on the twin assumptions that the knee behaves as a three rotational degree of freedom joint (no translations allowed) and that a model of STA could be developed from the DSX data in order to improve the knee joint angle estimates. This STA model was based on the location of the lateral knee marker relative to the knee joint center. It is important to note that although this method used a purely rotational knee model for the prior, the overall Bayesian cost function does not constrain the final knee motion to be rotation-only; it merely guides the solution to that suggestion. This is because our Bayesian likelihood, which is also part of the cost function, allows five degrees of freedom at the knee (three rotational degrees of freedom plus anterior/posterior and superior/inferior translation). The results of the example study revealed that the 6-DOF approach produced a poor Pose estimation compared to the DSX data (Fig. 1a, Table 1). This was, at least in part, because the three markers used to track the shank segment (on the lateral femoral epicondyle, the lateral aspect of the tibia, and the lateral ankle) were almost collinear (as is common for the conventional gait marker set). This near collinearity caused the 6-DOF solution to be hypersensitive to small measurement errors and STA. We next tested an IK model that had constrained the knee joint to permit no mediolateral translations. Thus the knee was, identically to the Bayesian likelihood, a 5-DOF joint (three rotations, plus anterior/posterior and inferior/superior translations), which permitted knee translations in the sagittal plane (Fig. 1b, Table 1). This model produced a poorer Pose estimation than 6-DOF for the given marker set when compared to the DSX results.
Fig. 1 A comparison of the DSX data (gold thigh and shank) and motion capture data (white thigh and shank) for an indicative sample frame. (a) When segment optimization (6 DOF) Pose estimation was used to track the thigh and shank. (b) When inverse kinematics was used to track the thigh and shank with a five degree of freedom knee constraint was used
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
189
Table 1 Test results (RMS error) of the tibia and femoral origin locations (motion capture vs. DSX) over all frames measured by the X-ray system Pose method 6 DOF Inverse kinematics (5 DOF) Inverse kinematics (3 DOF) Bayesian with motion and soft tissue prior
Left thigh error (in mm) 15.6 12.9 9.7 10.6
Left shank error (in mm) 32.0 41.9 13.2 4.4
Fig. 2 A comparison of the dynamic stereo X-ray data (gold thigh and shank) and motion capture data (white thigh and shank) for an indicative sample frame tracking the thigh and shank. (a) A three rotational degree of freedom inverse kinematic knee model was used to track the segments from the motion capture data. (b) Bayesian inference, with a prior that included a three degree of freedom knee and a soft tissue artifact model obtained from DSX data, was used to tack the segments from the motion capture data
The third and final test of the deterministic algorithms compared the DSX results to those when using an IK model with a 3-DOF knee that was constrained to permit no translation but allow rotations in all three planes. This produced the best non-probabilistic inference match between the motion capture data and the dynamic stereo X-ray data (Fig. 2a). However, a 3-DOF IK will never, by definition, allow any knee translation to be measured. Thus it is limited as a long-term solution to our Pose estimation problem, especially in a clinical setting, when joint translation may be a critical piece of information. For our probabilistic approach, we implemented a prior based on two assumptions: a 3 (rotational)-DOF knee joint and an STA model. The STA model, which is discussed in the next section, was based on a simple linear relationship between the axial rotation of the hip joint and the soft tissue motion of the lateral knee marker. It is important to reemphasize that the assumption of a 3-DOF knee prevented joint translation in the prior (second term on the right-hand side in Eq. 17); however, perhaps counterintuitively, the final overall solution (minimization of Eq. 17) still allowed translation at the knee joint because our Bayesian likelihood (the first term on the right side of Eq. 17) was based on a 5-DOF knee. The end result was a probabilistic Pose estimation that provided superior shank translational agreement
190
T.M. Kepple and A.R. De Asha A/P Knee Translation: X-Ray (black) and Bayesian (red)
A/P Translation (meters)
0.0000
−0.0050
−0.0100 0.0
50.0 % DSX Interval
100.0
Fig. 3 Comparison of the anterior/posterior knee translation as measured by the DSX system (black) and from the MoCap data after the Bayesian model was applied (red)
than the discriminative models and equivalent agreement for thigh translations (Fig. 2b and Table 1). To further demonstrate the potential of using a STA model without limiting joint translation, we can examine the data for the anterior/posterior knee translation (Fig. 3) as measured by the DSX system (black) compared to the anterior/posterior knee translation obtained from a Bayesian model (red). Note the translational displacement between the two models match up very well over the first 75% of the data interval. To develop our prior used in this model, we collected data on two subjects during walking at each’s self-selected walking speed on a treadmill. We then tested the algorithm on a third subject during the same task. We modeled the soft tissue motion of the lateral knee marker because the IK solution proved to be most sensitive to “accuracy” of this marker. Upon examination of the DSX and MoCap data, we identified a relationship between the soft tissue error of the lateral knee marker and the internal/external rotation of the hip (Fig. 4). When looking at the data from our initial two subjects, we found that the linear relationship between internal/external rotation angle versus the anterior/posterior soft tissue motion had an average r2 = 0.78 with an average slope of 1.0 mm/degree of rotation. Unfortunately, due to the limited amount of DSX data available, we did not find a strong enough relationship to predict vertical soft tissue motion and vertical knee translation. From Fig. 3, it is clear that although our prior included the expectation that the knee is a three rotational-DOF joint, the overall solution to the Bayesian approach allowed translation due to the inclusion of knee translation in the likelihood. We can
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
191
LHip Internal/External Rotation
degrees
7.4
2.1
−3.2 0.0
50.0
100.0
Ap / Soft Tissue (meters)
LKNE Targer A/P soft-tissue motion 0.00
−0.00
−0.01 0.0
50.0
100.0
% DSX Interval
Fig. 4 Plot of the hip internal/external rotation angle (top) versus the anterior/posterior soft tissue motion for the lateral knee marker (bottom) from one of the subjects
also observe from Fig. 3 that, for the first 75% of the X-ray-viewing interval, the Bayesian approach did a reasonably good job of predicting the magnitude and direction of knee translation. During the last 25% of the interval, the Bayesian approach did not match the DSX-measured knee translation well. This was because there was a weak soft tissue relationship described by our simple regression equation during the latter part of the viewing interval (Fig. 4). Ideally, during this latter part of the DSX interval, the prior should have been weighted less. Examining the DSX data for the two subjects used to develop the soft tissue model, we found that the soft tissue relationship was variable during the latter part of the DSX interval. Thus it could be postulated that, with enough DSX data to generate a motion prior that included more robust covariance matrices, our results would be improved during this period. These data represent only a proof of concept; a much larger sample size for generating the covariance matrices would improve the results further. Figures 3 and 4 illustrate the potential power of the probabilistic algorithm. From this sample work, we were able to demonstrate that a probabilistic Pose estimation approach greatly reduced translation estimation errors compared to traditional deterministic approaches (Table 1) while still allowing a good precision for our knee translation estimates throughout the first three quarters of the DSX interval (Fig. 3). We realize that developing better soft tissue models and generating improved
192
T.M. Kepple and A.R. De Asha
covariance matrices to properly weight the Bayesian terms at the appropriate times during a movement could further improve probabilistic Pose estimates.
Casting the 6-DOF and IK Methods in a Statistical/Generative Framework It should be noted that both the 6-DOF and IK deterministic methods for estimating model Pose may also be cast within a statistical framework. Although they both implicitly define a discriminative model and attempt to maximize the probability of the data under their respective model, we can restate these methods within a statistical/generative framework. This will hopefully allow the reader to better able to appreciate the relationship between the deterministic Pose estimations of the previous chapter and the probabilistic Pose estimation described in this chapter. To understand how both the 6-DOF and IK methods are a subset of our Bayesian optimization, let’s take another look at Eq. 17 above: f ðqÞ ¼ r ðq, vÞT V 1 r ðq, vÞ þ ðq ^ q ÞT U 1 ðq ^q Þ
(18)
This is the cost function for our probabilistic Pose estimation optimization problem. In this equation, r(q, v) is the residual of the fit between the measured and expected target locations, and ðq ^q Þ is the difference between the current generalized coordinate, q, and the expected generalized coordinate ^q, based on prior knowledge. V1 is the inverted covariance matrix for the X, Y, Z components of the residual, and U1 is the inverted covariance matrix for the difference between the measured and expected generalized coordinates. Now, in the 6-DOF method, there are no prior expectations of the values of the generalized coordinates. In the inverse kinematics method, the only prior knowledge we apply to the joint motion is that the motion becomes fully constrained. The net effect is that, for both the 6-DOF and IK methods, U1 becomes a zero matrix and the cost function reduces to: f ðqÞ ¼ r ðq, vÞT V 1 r ðq, vÞ
(19)
Another underlying assumption of both the basic 6-DOF and IK methods is that the residuals at any given frame and for any given target are weighted equally, which is theoretically equivalent to setting the covariance matrix V1 to the identity matrix. This would further reduce the cost function to: f ðqÞ ¼ r ðq, vÞT r ðq, vÞ
(20)
Examination of this cost function reveals that it represents nothing more than a minimization of the sum of squares error between the expected and measured target
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
193
locations. This is the equivalent of the cost function for an unweighted 6 DOF where the model Pose, q, is contained in the rotation matrix, T, and translation vector, O: r ðq, vÞ ¼
XN i¼1
ðvi Rai OÞ2
(21)
and an unweighted inverse kinematics problem: r ðq, vÞ ¼
XN i¼1
ðvi RðqÞai OðqÞÞ2
(22)
This derivation demonstrates that the deterministic solution is simply a subset of the generative solutions to the estimation of Pose.
Future Directions In this chapter, we have highlighted our initial success at using a model of soft tissue artifact as a prior. Although the simple STA model used in our test example worked reasonably well for our simple test example, establishing a general prior for STA that will be applicable to different marker sets, patient populations, and movement patterns is unlikely. In other words, establishing priors for STA and other expectations of the Pose will be quite challenging. Thus, in our view, it is imperative that probabilistic Pose estimation evolve in a manner that will facilitate the development of custom Bayesian priors. In addition to dealing with STA, custom priors will allow probabilistic Pose estimation to aid new motion capture technologies, such as markerless motion capture (▶ “3D Dynamic Pose Estimation from Markerless Optical Data”), to move into clinical and research settings. Note that once a new prior is created, a new set of covariance matrices is required for use with that prior. When insufficient data are available to generate the covariance matrices from a large sample, the approach can still be used by either using a smaller sample size and only the diagonals of the covariance matrices or simply by weighting the matrices so that the likelihood and prior terms are equal. It is expected that diagonal covariance matrices or ad hoc weighting will likely diminish the accuracy of the probabilistic method but will still provide improved Pose estimations better than discriminative methods. One statistical technique that might possibly allow for adapting priors, in order to facilitate their being generalized across populations (e.g., a patient group), is waveform principal component analysis (PCA) (Deluzio and Astephen 2007). PCA can be used to describe a normalized waveform (e.g., one of the generalized coordinates describing the Pose). Given a mean signal of a generalized coordinate for a motion (Pose being described by this collection of signals), and a small number of principal components, a signal may be constructed that could approximate signals from all subjects in the control data set (subject-specific signals differ by a set of coefficients
194
T.M. Kepple and A.R. De Asha
or PCA scores). Consider now a rich data set that includes multi-subject metadata such as height, weight, age, sex, and disease or condition, as well as the condition severity. In addition to representing subject-specific signals, PCA can be used to discriminate or differentiate groups; therefore the shape of the prior could, potentially, be predicted using these metadata and the principal components alone. The power in this method is that it is predictive and could be calculated prior to collection. Accordingly, it could be used as a Bayesian prior for estimating Pose. This approach would also have the advantage that it would always have a prior, even in the absence of data, and thus it would provide a solution (e.g., the prior), rather than no solution, in situations when the model is not observable. In a deterministic solution, as described in the previous chapter, an unobservable system state results in the entire frame of data being unreliable and therefore “empty.” In the future, it should be possible to expand Bayesian priors to include multibody dynamics via optimal control. Treating multibody Pose estimation as an optimal control problem was described earlier in this chapter and is well established. The objective is usually to determine controls that produce a movement that is, in some sense, “optimal.” Optimal control, via direct collocation (DC) methods (van den Bogert et al. 2011; Miller and Hamill 2015) (▶ “Optimal Control Modeling of Human Movement”), casts Pose estimation as a trade-off between an initial estimate of the Pose from the MoCap data, multibody dynamics, and from matching the recorded ground reaction force. In the future, we can envisage use of the output of the DC solution as a Bayesian prior that would, in essence, drive the Pose estimation from the marker based MoCap data toward dynamic consistency in a controlled way. This would represent a “full circle,” as the use of multibody dynamics-based priors was the starting point for this probabilistic approach (Todorov 2007), and has already been extended by Todorov’s lab for robotics (Lowrey et al. 2017) and may be the most likely path of future progression for human motion.
References Anderst W, Zauel R, Bishop J, Demps E, Tashman S (2009) Validation of three-dimensional model based tibio-femoral tracking during running. Med Eng Phys 31(1):10–16 Andriacchi TP, Alexander EJ, Toney MK, Dyrby C, Sum J (1998) A point cluster method for in vivo motion analysis: applied to a study of knee kinematics. J Biomech Eng 120:743–749 Cappello A, Stagni R, Fantozzi S, Leardini A (2005) Soft tissue artifact compensation in knee kinematics by double anatomical landmark calibration: performance of a novel method during selected motor tasks. IEEE Trans Biomed Eng 52:992–998 Davis R, Ounpuu S, Tyburski D, Gage J (1991) A gait analysis data collection and reduction technique. Hum Mov Sci 10:575–587 Deluzio KJ, Astephen JL (2007) Biomechanical features of gait waveform data associated with knee osteoarthritis. An application of principal component analysis. Gait Posture 25:86–93. PMID: 16567093 Dumas R, Camomilla V, Bonci T, Cheze L, Cappozzo A (2014) Generalized mathematical representation of the soft tissue artefact. J Biomech 47:476–481
3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras. . .
195
Grimpampi E, Camomilla V, Cereatti A, de Leva P, Cappozzo A (2014) Metrics for describing softtissue artefact and its effect on Pose, size, and shape of marker clusters. IEEE Trans Biomed Eng 61(2):362–367 Hamner S, John C, Anderson FC, Higginson J, Delp S (2008) Reducing residual forces and moments in a three-dimensional simulation of running. Proceedings of the North American Congress on Biomechanics IV, August 2008 Kadaba M, Ramakrishnan H, Wootten M, Gainey J, Gorton G, Cochran G (1989) Repeatability of kinematic, kinetics and electromyographic data in normal adult gait. J Orthop Res 7:849–860 Kaplan ML, Heegaard JH (2001) Predictive algorithms for neuromuscular control of human locomotion. J Biomech 34:1077–1083. PMID: 11448699 Koelewijin A, Richter H, van den Bogert A (2016) Trajectory optimization in stochastic multibody systems using direct collocation. In Proceedings of the 4th joint international conference on multibody system dynamics Leardini A, Chiari L, DellaCroce U, Cappozzo A (2005) Human movement analysis using stereophotogrammetry. Part3. Soft tissue artifact assessment and compensation. Gait Posture 21:212–225 Lowrey K, Dao J, Todorov E (2017) Real-time state estimation with whole-body multi-contact dynamics: a modified UKF approach. In: Humanoid Robots (Humanoids), 2016, IEEE-RAS, 16th International Conference Lu TW, O’Connor JJ (1999) Bone position estimation from skin marker co-ordinates using global optimization with joint constraints. J Biomech 32:129–134 Miller R, Hamill J (2015) Optimal footfall patterns for cost minimization in running. J Biomech 48:2858–2864 Miller R, Kepple T, Selbie WS (2016) Direct collocation as a filter for inverse dynamics. In Proceedings of the 40th annual meeting of the American Society of Biomechanics Peters A, Galna B, Sangeux M, Morris M, Baker R (2010) Quantification of soft tissue artifact in lower limb human motion analysis: a systematic review. Gait Posture 31:1–8 Remey C, Thelen D (2009) Optimal estimation of dynamically consistent kinematics and kinetics for dynamic simulation of gait. ASME J Biomech Eng 13(3):31005 Spoor C, Veldpaus F (1980) Rigid body motion calculated from spatial coordinates of markers. J Biomech 13(4):391–393 Stroupe AW, Martin MC, Tucker B (2001) Distributed sensor fusion for object position estimation by multi-robot systems. In IEEE international conference on robotics & automation (ICRA-01) Taylor WR, Ehrig RM, Duda GN, Schell H, Seebeck P, Heller MO (2005) On the influence of soft tissue coverage in the determination of bone kinematics using skin markers. J Orthop Res 23:726–734 Todorov E (2007) Probabilistic inference of multijoint movements, skeletal parameters and marker attachment from diverse motion capture data. IEEE Trans Biomed Eng 54:1927–1939 van den Bogert A, Blana D, Heinrich D (2011) Implicit methods for efficient musculoskeletal simulation and optimal control. Procedia IUTAM 2(2011):297–316
3D Dynamic Pose Estimation from Markerless Optical Data Steven Cadavid and W. Scott Selbie
Abstract
This chapter provides an overview of three-dimensional (3D) dynamic Pose (position and orientation) estimation of human movement without the use of markers or sensors, more commonly known as Markerless Motion Capture (Markerless Mocap). As with Marker-based Motion Capture (Marker-based Mocap), the methods presented estimate the Pose of an underlying multibody subject-specific model comprising rigid segments with anatomically defined local reference frames and joint constraints. In addition, the model has an overlying surface representing the skin, or clothing, depending on the context. The focus of this chapter is on Markerless Mocap algorithms best suited to biomechanical analyses of human movement. In other words, those techniques appropriate for estimating 3D Pose directly, and accurately, from recorded data. Of all the approaches to Markerless Mocap, 3D-to-3D Pose estimation is most similar to Marker-based Mocap techniques because it requires arrays of multiple, time synchronous, video cameras encircling the capture volume. In addition to the underlying multibody skeletal model that marker-based and markerless techniques have in common, during Markerless Mocap, the subject is identified by a surface model overlying the skeleton. In each frame of motion data, a pixelated surface, comprised of a dense collection of points lying on the surface, is extracted from the scene and registered to the model. Neither marker-based nor 3D-to-3D Markerless Mocap is typically accurate enough to record the Pose of the bones at a resolution for studying joint dynamics. S. Cadavid (*) KinaTrax Inc., Palm Beach, FL, USA e-mail: [email protected] W.S. Selbie HAS-Motion Inc, Kingston, ON, Canada C-Motion Inc., Germantown, MD, USA e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_160
197
198
S. Cadavid and W.S. Selbie
An alternative markerless approach to joint level biomechanics has emerged. Biplanar videogradiography (or Dynamic Stereo X-ray) uses a 3D-to-2D approach to Markerless Mocap, whereby only two views of the subject are acquired because of space limitations and to minimize radiation exposure. A brief introduction to 3D-to-2D registration will be presented because this is covered in more detail in another chapter. Keywords
Markerless Mocap • Marker-based Mocap • Multibody 3D Pose estimation • Articulated registration • Space carving • Stereo reconstruction • Biplanar videoradiography • 3D-to-3D registration • 3D-to-2D registration • Visual hull
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a Visual Hull Representation of the Surface of the Subject . . . . . . . . . . . . . . . . . . . . . . Modeling the Surface from Dense Voxelization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Identifying Subject-Specific Reference Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tracking the Pose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3D-2D Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
198 199 201 205 206 208 210 212 214 216 216
Introduction There are numerous challenges when recording human movement using markers or sensors attached to a subject’s skin or clothing. Data collection protocols for threedimensional Marker-based Mocap are time-consuming and expensive and require a constrained laboratory/studio environment. Marker-based protocols require the individual collecting data to have a high degree of technical expertise, as segmental Pose (position and orientation) estimation can be extremely sensitive to marker placement precision (Della Croce et al. 2005). In common with all Marker/sensor-based Mocap systems, the encumbrance of the markers or sensors and the instructions to participants, to behave “naturally” on demand, can cause many subjects to act “unnaturally.” In this chapter, the exemplar motion capture being illustrated is the recording of a Major League Baseball pitcher throwing during live game action. In this game scenario, the Mocap protocol must not affect the game or the pitcher, and all players must be unaware of the recording. Lastly, and in some respects most importantly, markers and sensors move relative to the underlying skeleton to which they are attached (Cappozzo et al. 1996). This soft tissue artifact is idiosyncratic and challenging to mitigate mathematically. Markerless Mocap requires no markers or sensors and thus reduces the time and expertise required to collect data. Markerless Mocap allows subjects to move
3D Dynamic Pose Estimation from Markerless Optical Data
199
naturally through a motion capture volume that may be indoors or outdoors. Because there are no errors associated with marker/sensor placement, Pose estimation is dependent, predominantly, on the mathematical algorithms used for processing the data. These can always be refined as mathematical techniques improve, so no data are lost to operator expertise. Markerless Mocap therefore has the potential to increase the availability and reduce the cost, of performing accurate 3D movement analysis. The ability to collect reliable motion capture data efficiently and independently of the operator facilitates the collection of large amounts of data across a broad spectrum of environments. Markerless Tracking is the process for estimating the Pose of segments that comprise a model of a subject, in each frame of a recorded sequence. This chapter focuses on optimization-based approaches to Pose estimation including methods that register a 3D articulated multibody model to the surface of the subject (3D-to-3D registration; Corazza et al. 2010; Cheung et al. 2003), and methods that generate multiple virtual two dimensional (2D) images (digitally reconstructed video image) and register these 2D images to recorded video images (3D-to-2D registration; Stoll et al. 2011; Balan et al. 2007, Bey et al. 2006, 2008; Brainerd et al. 2010; Giphart et al. 2012; Haque et al. 2013; Tashman et al. 2017). Omitted from this chapter are methods of sensor fusion, including Markerless Mocap and IMU (Marcard et al. 2016), and methods predominantly focused on animation and vision, such as 2D-to3D Markerless Mocap from one video camera (Bogo et al. 2015), or estimations from sparse markers and statistical shape models (Loper et al. 2014).
State of the Art As in other chapters of this Handbook, the assumption underlying markerless Pose estimation is that the body (Fig. 1a) is comprised of a skeletal model (Fig. 1b) constructed from a set of rigid (nondeformable) segments (or bodies) and, unique to a markerless approach, a surface model (Fig. 1d) representing the skin and/or clothing. Each skeletal segment is defined by a local anatomical reference frame (right handed Cartesian coordinate system). The origin of the reference frame is placed at the proximal end of a segment, coincident with the distal end of an adjacent segment. This creates a joint connecting the child (usually the distal segment) to the parent (usually the proximal segment, Fig. 1b). Each segment is restricted to having only one parent segment, and the segment’s interaction with its parent segment is described by the specification of joint constraints acting at, and around, the origin of a segment, relative to the parent segment. These joint constraints define the number of degrees of freedom allowed at the joint. For Markerless Mocap, the number of degrees of freedom can be any integer value between zero and six. In practice, however, only the root segment is prescribed 6 DOF relative to the laboratory, and all other segments have between 1 and 3 DOF relative to the parent, i.e., segments are allowed to rotate relative to each other, but translation tends to be constrained. For marker-based Mocap, the local reference frames defining the segments of the multibody model are typically generated from the location of markers placed on
200
S. Cadavid and W.S. Selbie
Fig. 1 (a) The subject is defined by an underlying skeleton and surface. (b) For the baseball example in this chapter the skeleton is represented by a set of geometric primitives linked as a multibody model. (c) An example of segment anatomical reference frames typical of Marker-based Mocap. (d) The surface of the body is defined by a polygonal mesh rigged to the skeleton. An exemplar frame showing the skeleton and surface model (Visual Hull) of a pitcher
several anatomically defined locations (palpable bony landmarks) in a reference trial, or by the relative motion of sets of homologous landmarks markers in a functional reference trial (Fig. 1c). With respect to the local segmental reference frames, the Pose of a template is defined by a set of tracking markers attached to the segments. In each frame of motion data, this template is registered to the 3D location of homologous markers from the recorded scene. From this registration, the Pose of the multibody model is computed. Similarly, a 3D-to-3D markerless Mocap solution identifies the locations of a set of surface points with respect to local segment reference frames. The articulated skeleton is an underlying structure that defines the Pose of the subject, but because it is hidden within the body, it does not provide data needed for the markerless Pose estimation. An independent approximation to the subject’s
3D Dynamic Pose Estimation from Markerless Optical Data
201
skin/clothing is established as a polygonal surface model that is rigged to the underlying skeleton (Fig. 1c). In order to represent a specific subject, this skeletal/surface model must be scaled and transformed to match the size and shape of the subject. Instead of discrete identified markers, however, markerless Mocap uses a dense set of unidentified 3D points representing an entire pixelated surface of the subject (Template Point Cloud). All 3D-to-3D methods are based on identifying a cloud of 3D points that define the surface of the subject (a Visual Hull) and map this surface to an underlying representation of the skeleton. In each frame of motion data, the Template Point Cloud is segmented from the scene and registered to the Visual Hull. Instead of the one-to-one mapping used in a marker-based approach, however, the two point clouds are registered as a collection (3D-to-3D registration). From this registration, the Pose of the multibody model is computed. Of the many techniques described in the literature to recover the 3D surface shape of an object from multiple calibrated viewpoints; In this chapter, Space Carving (Kutulakos and Seitz 2000) will be presented in detail, and Stereo Reconstruction (Seitz et al. 2006) and depth sensors (Weiss et al. 2011) in summary.
Creating a Visual Hull Representation of the Surface of the Subject Of the many approaches to markerless Mocap, 3D-to-3D Pose estimation is most similar to marker-based Mocap techniques because it requires arrays of multiple time synchronous video cameras encircling the capture volume. As with marker-based Mocap, the number and configuration of cameras is dependent on the type of motion to be captured and the size of the capture volume. The camera lenses and calibrated volume are adjusted such that every camera’s field of view contains the movement being analyzed with as wide a variety of unobstructed views of the movement as possible. In the Major League Baseball parks, arrays of 7 to 16 cameras are mounted to the exterior structure of the stadiums (Fig. 2). Space carving is a method for identifying the Visual Hull of an object from projections of a silhouette of the subject (in the example of this chapter, the baseball pitcher) onto multiple cameras. This 3D-to-3D Pose estimation is often referred to as a “silhouette based” approach. The prerequisite, therefore, is to extract a silhouette of the subject from the background. In order to identify pixels in a video that correspond solely to the subject (in this example, the pitcher), a statistical model of the background (all nonpitcher pixels) must be created and removed from each video image. Of the many ways to model the background (Piccardi 2004), the simplest approach is to use a single image of the capture volume prior to the subject entering the scene. The limitation of this approach is that the background may vary during the pitch resulting in background pixels that substantively change in appearance being detected as foreground pixels (i.e., as part of the subject). Background models that can cope with changes in illumination conditions and appearance are typically statistical methods, such as the Gaussian Mixture Model (Zivkovic 2004), which compute a multimodal distribution at each pixel location across multiple exemplar background images. For a series of background images, each pixel is assigned a likelihood (of being a background pixel) based on its intensity and the respective
202
S. Cadavid and W.S. Selbie
Fig. 2 A schematic drawing of a baseball stadium with 7 cameras aimed at the pitcher’s mound. The figure displays the image of a pitcher during the delivery of the ball from each of the 7 cameras. At the bottom of the image is a controlling computer synchronizing the cameras and receiving the video data
distribution of intensities of every other pixel in the image. Incorporating multiple background images, with varying appearances (e.g. at different times during the game), to build pixel intensity distributions enables the background model to be more robust to changes in background appearance during the data collection. The silhouette of the subject is isolated by subtracting the background and thresholding the resultant difference image to yield a binary image for each video image frame. Holes and rough edges in the silhouette, where a background pixel was inadvertently identified, are corrected through a series of binary morphological operations that fill in holes and smooth the resulting silhouette (Fig. 3). For Space Carving, each voxel (volume pixel) is projcted onto the image plane of a camera; all voxels that do not intersect with the silhouette are assigned a value of zero (e.g., carved out of the voxel array). This process is repeated for each camera view. The resulting nonzero voxels represent a carved representation (Convex Hull) of the subject for that frame of data. Space Carving is robust to false positives in the foreground detection (i.e., background pixels incorrectly labeled as foreground pixels) because the resultant Visual Hull is the intersection of the back-projections and foreground. Note that the resulting shape cannot recover the smoothness and concave surfaces of the subject because many of these features lie within the silhouette. If, however, these “enclosed” surfaces have voxels with a unique color from the rest of the silhouette, the silhouette can be assigned multiple colors and a color consistency check may be used to carve out these regions independently (Kutulakos and Seitz 2000).
3D Dynamic Pose Estimation from Markerless Optical Data
203
Fig. 3 A silhouette of a baseball pitcher at a frame in the middle of the throwing motion extracted from a video image
Fig. 4 Two views of a pitcher at different time points in the pitch carved from multiple cameras. Note the problem created by shadows, and the concave regions (for example, the ankle) that have been filled by the space carving technique
A challenge facing the Space Carving technique, when collecting data in outdoor settings (as in our baseball example), is inconsistent lighting that casts shadows of the subject (Fig. 4). The simplest and most effective approach to remove shadows is to place a camera close to the ground with its optical axis parallel to the floor. This low-placed camera could “carve away” the cast shadow because it is not visible from this viewpoint, and hence discarded during the intersection of the silhouette backprojections (KaewTraKulPong and Bowden 2002). In our baseball example, however, cameras cannot be placed on the field, so the cast shadow must be identified by the Pose estimation as noise and ignored. The voxelization obtained by Space Carving is referred to as a dense Visual Hull voxelization because all surface and internal voxels are included in the Visual Hull.
204
S. Cadavid and W.S. Selbie
This dense voxelization then undergoes shelling and water-tight meshing, using the Marching Cubes algorithm, to obtain a Visual Hull. The surface of the hull, a thinshelled voxelization, is produced by applying binary morphological cleaning to the dense voxelization. This thin-shelled Visual Hull voxelization is an approximation to the subject’s skin (Fig. 4). An alternative approach to generating the Visual Hull is Stereo Reconstruction. In this approach, a set of common fiducial (reference) landmarks are identified across adjacent cameras. Methods such as the Harris Corner Detector (Harris and Stephens 1988) and the FAST Corner Detector (Rosten and Drummond 2006) can be used to identify these fiducial landmarks. This is done within a small window of the entire image where a strong gradient, within that window, is present. For each fiducial window, a feature vector is generated to represent the texture within this local window surrounding a fiducial point. The feature vector is generated using rotation and scale invariant feature extraction methods such as SIFT (Lowe 1999), SURF (Bay et al. 2006), and ORB (Rublee et al. 2011). Candidate keypoint correspondences are then established by pairing key points in two images that are most similar in the feature space. Outlier correspondences are filtered by aligning the key points in one image to their correspondences in the second image. This is achieved by applying an affine transformation that minimizes the root mean square distance error between the point sets. Corresponding key points that exceed a predefined distance threshold are then discarded. The optimal affine transformation used to align the correspondences can be computed using Procrustes Analysis (Gower 1975). This process yields a sparse set of keypoint correspondences between the two images, which alone is insufficient to generate a dense point cloud reconstruction of a subject. To address this, the images undergo rectification to transform them to a common image plane. This results in all epipolar lines being parallel to the horizontal axis and corresponding points having identical vertical coordinates. The problem of establishing dense point correspondences is then greatly simplified, because the correspondence search is reduced to a 1D search in the same row of the counterpart image. Simple and efficient pixel intensity comparisons can then be applied to identify a correspondence for each pixel in the two images. A disparity map (the apparent pixel difference or motion between a pair of stereo images) is then generated from the pixel location distances between correspondences. A disparity map can be transformed into a dense 3D point cloud by triangulation. The Stereo Reconstruction method can be extended to more than two views to form a complete 3D model by reconstructing a point cloud from every adjacent pair of cameras. Feature correspondences can also be established across more than two views to obtain a more accurate measure of depth using Bundle Adjustment (Triggs et al. 1999). Stereo Reconstruction can produce accurate 3D reconstructions recovering convex and concave surfaces; however, the primary disadvantage is that it relies on the veracity of point correspondences to produce an accurate result. Objects with little texture, such as skin and some articles of clothing, yield poor correspondences since the resultant feature vectors are insufficiently discriminative. Therefore, most researchers opt for Space Carving because it does not require the establishment of point correspondences.
3D Dynamic Pose Estimation from Markerless Optical Data
205
Lastly, and presented only briefly, active depth sensing technologies such as Structured Light Sensors and Time of Flight (ToF) sensors (e.g., Microsoft Kinect) can be used to generate a Visual Hull. Depth sensors project a constant pattern of infrared (IR) light (dots) into the capture volume. As the light strikes, and is reflected from a surface, the pattern is distorted. This distortion is recorded by depth cameras located at an offset from the IR transmitter. From the difference between the recorded and expected dot positions, the depth of each pixel of an RGB (red, green, blue) color camera is computed. Perhaps the best known implementation of this approach is the Microsoft Kinect. The Kinect algorithm (Shotton et al. 2013; Keskin et al. 2013) extracts an efficient feature vector from each pixel in a depth image and classifies the pixel as belonging to a specific body part. To create a complete Visual Hull from depth sensors, multiple sensors are typically used in which the depth map identifies pixels on the surface of the subject. These pixels are converted into 3D points on the surface of the object (i.e., the Visual Hull). To cover the entire surface of the object, the 3D points captured by multiple depth cameras from different viewpoints can be integrated. Points on background objects (i.e., objects except for the subject) are excluded typically by the difference in depth from the subject. This approach tends to be useful in a small volume because the depth sensors have only a short range. In the case of the baseball example in this chapter, in which there is no possibility of getting close to the pitcher at any time, this approach is not appropriate.
Modeling the Surface from Dense Voxelization The next step in this markerless approach is to model this dense point cloud of data as a mathematical surface. In this chapter, two types of subject-specific body surface models are presented: generic surface templates and reconstructed Visual Hull models. Generic surface template models consist of a generic skin surface that, by default, represents a parameterized body shape. Typically the model also contains an underlying generic skeleton that is rigged to the skin surface. A simple parameterized segment surface can be represented by a small series of spheres along the axis between joints, as used successfully to track non human primates (Nakamura et al. 2016) This model must be transformed to register with the size and shape of the subject. This model is perhaps best represented by a Statistical Shape Model (Anguelov et al. 2005; Pons-Moll et al. 2015) that is derived from a collection of similar subjects with the surface synthesized from a weighted set of shape modes of variation. The shape modes of variation are computed by applying principal component analysis to a dataset of varying body morphologies. The skin surface consists of a mesh, where each skin vertex on the mesh, p, is assigned a set of weights, wi. These weights control the amount of deformation that a vertex can undergo as a result of n neighboring segments being rotated during registration:
206
S. Cadavid and W.S. Selbie
pi ¼
n X i¼1
w i Ri p
n X
wi ¼ 1
(1)
i¼1
where Ri corresponds to the rotation matrix applied to the ith neighboring segment and pi signifies the transformed skin vertex position. In a process that is commonly referred to as skinning, the generic model is made subject-specific by applying a nonrigid registration between the generic model and a 3D representation of the subject (i.e., the Visual Hull described in the previous section) that has been captured from the acquisition sensors. This registration process is challenging because, at this stage, the subject’s segment lengths, body morphology, and the texture information for the subject are all unknown. When registered well, the generic template models tend to produce a clean smooth skin surface and underlying skeleton because the model is usually handmade by a graphic artist or animator. However, the registration optimization procedure is often ill-constrained and the resultant skin surface may not fully conform to the actual morphology of the subject. Reconstructed Visual Hull models can be obtained by a 3D scan of the body (Corazza et al. 2010) or can be constructed directly from the Visual Hull extracted from the subject performing the movement of interested (e.g., representing many of the Poses inherent to the motion). Space Carving is typically used to reconstruct the texture-mapped point cloud of the subject’s skin. A skin mesh is then constructed from the point cloud using the Marching Cubes algorithm (Lorensen and Cline 1987). Because Visual Hull models are directly constructed from the actual morphology of the subject, the skin surface may contain artifacts due to the limitations of the Space Carving procedure.
Identifying Subject-Specific Reference Frames Given the surface approximation of the subject, the next step is to identify the Pose of a skeleton within this surface. To construct the underlying skeleton, joint centers (i.e., the origin of each segment anatomical reference frame) must be identified within the Visual Hull. The lengths of the segments in a multibody model can often be measured or identified from selected postures. In the baseball example, the data are collected during a live Major League baseball game, and it is essential that the pitchers are not interfered with (even before the game) and that the game itself is not interfered with in any way. Therefore, recording a predefined A-Pose or T-Pose is not practicable; instead, the joint centers are digitized manually from video images of a recorded pitch. The carved Visual Hull deforms during the pitch, so several key frames are digitized. At each keyframe, the joints of the subject are annotated (digitized) manually from two or more camera views (Fig. 5). A subject-specific multibody model with joint-specific constraints suitable for pitching is derived from a collection of the 3D location of joint centers (Fig. 5). Identifying subjects’ joint centers during a baseball pitch requires multiple
3D Dynamic Pose Estimation from Markerless Optical Data
207
Fig. 5 Joint centers as digitized manually, and displayed as colored circles
keyframes to account for deformation of the Visual Hull at different times during the pitch; thus, joint center locations vary between frames. A mean joint center position is estimated using a Sparse Bundle Adjustment algorithm (Lourakis and Argyros 2009). Each segment of a template skeleton, at each digitized video frame, is scaled to the distance between its proximal and distal joint centers. This collection of joint centers is then registered to each other using Procrustes Analysis. Using an iterative scheme, an optimal Pose of the constrained model, which minimizes the distance between the descendants of a given joint (distal segments) over all video images, is estimated. The lengths of the segments are computed as the mean lengths, calculated across the keyframes, and are used in the skeletal model for all subsequent processing. The resulting underlying skeleton, therefore, consists of rigid segments connected by joints. Given constant segment lengths and joint constraints, it is possible that for some Poses of the model during a movement, the computed average joint centers may not be anatomically congruent, relative to the surface. For example, a large asymmetry in the side-to-side anthropometrics or anatomically inconsistent joint placement (e.g., right and left shoulder joints are “flipped”). If an error is detected, the 2D annotations are adjusted manually and the average joint centers are recomputed. This process is repeated iteratively until an anatomically credible 3D joint center is obtained over multiple pitches. Given the multibody skeleton and Visual Hull, the individual voxels comprising the Hull must be categorized with respect to the skeleton. Depth Map Classification-based methods (Shotton et al. 2013) can be used to assign each pixel in the image to a specific segment of the skeleton. The classification features are the set of distances in depth between groups of two pixels at predetermined offsets from the pixel being classified. Pixels corresponding to each body part are clustered and fit to a skeletal model on a frame-by-frame basis. The classifier is usually trained on a large set of human movement Mocap data, in which pixels associated with each segment are identified manually. Given this training set, machine learning algorithms are typically used to estimate the Pose of the multibody model from the depth map and the association between a pixel and a segment. In this way, a skeletal model can
208
S. Cadavid and W.S. Selbie
be extracted from the depth map. In other words, each voxel must be assigned to one or two of the underlying nearest skeleton segments. Skin weights are computed for a given skin vertex by determining the closest segment to the vertex using point-to-line segment distance. The skin vertex is assigned to the nearest segment, and a secondary weight is computed for the nearest segment that is connected to the assigned segment. The weights are computed by conducting a k nearest neighbor search at the location of the skin vertex. The weight associated with a segment is equal to the fraction of nearest neighbors that are assigned to the segment. The value of k determines the extent of smoothing during the skin deformation. These weights dictate the deformation that a skin vertex undergoes as the model moves. Appropriate skin weights ensure smooth deformation of the skin around the joint centers during articulation. Visual Hull models constructed across multiple, varying Poses of a subject can improve the quality of both the skin surface reconstruction and the accuracy of the underlying skeletal Pose estimate. Given the segment lengths and locations of the joint centers, the alignment of the anatomical reference frames (AFs) must be computed. Unlike marker-based systems in which anatomical landmarks can be identified, the Visual Hull is not sufficiently unique to be considered in a reference alignment. The surface model and Visual Hull are globally aligned by applying a similarity transformation to the surface model, such that the centroids and first two principal components of both models coincide. The first two principal components, corresponding to the two largest eigenvalues, are computed by applying Singular Value Decomposition to the surface hulls. The models are scaled such that the distance between the centroids and their respective farthest point cloud neighbors is the same. A nonlinear registration process is used to optimally align each body region of the model to the surface hull. To obtain an optimal alignment, each body region of the model undergoes a nonrigid transformation comprising a rotation and nonuniform scaling about the axes of its anatomical reference frame (AF) while maintaining symmetry in the side-to-side anthropometrics. The resulting alignment defines the AFs for the model.
Pose Estimation At each frame of a movement trial, the Visual Hull of the subject is segmented from the background using the same methods described earlier in this chapter. Pose estimation is the process for registering the multibody skeleton and surface model to the Visual Hull. 3D-to-3D registration methods commonly use a variant of the Iterative Closest Point algorithm (Besl and McKay 1992), with scaling capability, to register the point clouds of the 3D articulated model and 3D frame reconstruction on a segment-by-segment basis. It should be noted that additional, nonrigid transformations can be incorporated into the objective function beyond affine scaling, to better conform to the subject body morphology. This, however, is at the cost of having additional variables in the optimization problem.
3D Dynamic Pose Estimation from Markerless Optical Data
209
At all frames of the movement, the Pose of the subject specific skeletal model is estimated. Using the same background subtraction and space carving algorithms as in the model identification, the voxels (points) comprising the subject are segmented, and from this dense voxelization, the surface voxels (point cloud) are identified (Note: for the movement trial the Visual Hull is not computed as only the point cloud is required for the Pose estimation). At each video frame, the Pose is estimated by registering the Visual Hull (rigged to the skeleton) to the surface point cloud. This is done using the articulated iterative closest point method. Notably, this does not require the correspondence between model vertices and the point cloud generated at a given frame, as global optimization methods are used to identify the optimal fit of the Pose in each frame. Seed Pose estimation involves identifying the subject in the scene and registering a multibody model to the subject in all video images from one frame. If possible, the frame of data selected for the Seed Pose is typically based on a frame of data in which the registration is considered straightforward on the basis of the subject being in a “known” or “predictable” Pose. Approaches to seed Pose detection differ depending on whether cooperation from the subject can be expected. In cooperative scenarios, a multibody model of the subject, including segment scaling, can be created by measuring the subject or by computing the location of joint centers functionally. For example, the subject could be instructed to assume a known position, such as the common T-Pose (subject standing upright with feet shoulderwidth apart and arms straight out to the sides, horizontal to the floor, palms facing forwards). The surface model is then configured in the known Pose and that is used as the initial seed Pose estimate. Virtual markers can also be manually placed at anatomical locations similar to Marker-based Mocap to help guide the nonrigid registration. For instance, a virtual marker at the elbow may be useful if the subject is in a T-Pose where the upper arm and forearm may be difficult to disambiguate due to lack of flexion at the elbow. The subject could be requested to wear specific clothing that is colored consistent with colors assigned to the Surface model. In this registration procedure, only the rotational degrees of freedom are retained as variables in the optimization provided fixed segment lengths are specified by the skeletal model. In uncooperative scenarios, where the subject is unaware of the motion capture, or where the subject cannot cooperate (e.g., during a competition), a known Pose may be detected within the trial. There are a variety of ways to detect a known Pose; however, one straight forward way in 3D-to-3D tracking methods is to construct the Visual Hull of the known Pose from a set of exemplar frames of the subject in that Pose. The iterative closest point algorithm can be applied between the Visual Hull Pose template and the 3D reconstruction obtained at each frame of the motion trial. Frame instances that yield residual alignment errors below a specified threshold are accepted as containing the known Pose. If a known Pose cannot be detected, the method of last resort is to manually digitize the joint centers in all of the video images of one frame (see Fig. 5).
210
S. Cadavid and W.S. Selbie
Tracking the Pose Once a Pose estimate has been computed for the initial frame, the tracking procedure is applied for each subsequent frame. The Pose estimate computed for a frame instance, t, can be used as the initial seed to compute the estimates at frame instances t 1 (backwards tracking) and t + 1 ( forward tracking). Backwards and forward tracking can be performed independently and concurrently to potentially reduce processing time. The concept of template matching to detect known Poses in the motion trial can be extended to also detect multiple key Poses in the motion trial. For instance, in baseball pitching, there are key Poses such as the setup, cocking phase, ball release, and follow through that must be present in each pitching sequence because they are important in the biomechanical analysis. These events can be detected and used as sample Poses to build a Pose interpolant. For instance, a spline interpolant can be built for each Euler angle component of each joint to estimate the trajectory of the joints. An interpolated Pose may result in joint angles that violate the joint constraints. Therefore, each joint in the interpolated Pose must be adjusted to ensure it conforms with the defined constraints. One example procedure to building the Pose interpolants is as follows. Consider the case where a key Pose is detected at frames 0 and n, n > 2. A Pose interpolant, γ 0, is built using the estimates at frames 0 and n to compute initial seed Pose estimates at frames 1 and n 1. At the following iteration, Pose interpolant, γ 1, is built using the Pose estimates at frames 0, 1, n 1, and n to compute initial seed Pose estimates at frames 2 and n 2. This process is repeated until all frames between 0 and n have a computed Pose estimate. In cases where a given frame to be tracked is not located between two key Posedetected frames, extrapolation can be used to refine the initial seed of the frame. To alleviate the computational bottleneck of Space Carving in 3D-to-3D tracking methods, the initial seed Pose estimate can be used to reduce the space carved volume to a subvolume within the capture volume enclosing the initial seed Pose estimate. The reduction of the space carved volume on a frame-by-frame basis will also reduce the effect of any nonsubject artifacts present in the Space Carving reconstruction. The articulated registration procedure employed in tracking can utilize local optimization techniques, provided a good initial seed Pose estimate is available. The optimization consists of only rotational degrees of freedom being allowed at each joint in addition to translational and rotational degrees of freedom at the root joint, in cases where fixed segment lengths across the Pose estimates of a motion trial are imposed. Constraints that are applied to the tracking optimization include joint constraints, displacement constraints that restrict the amount of rotation possible between adjacent frames, shape constraints, texture constraints, kinetic constraints, and surface collision constraints. In 3D-to-3D tracking methods, shape constraints can be imposed by computing surface normals at each skin vertex on the articulated model, as well as on the points of the 3D frame reconstruction. The registration procedure can account for the angular difference between the surface normals of registered points to ensure the registered shapes are similar. Articulated registration techniques typically apply segment-wise registration that follows the hierarchical
3D Dynamic Pose Estimation from Markerless Optical Data
211
order of the kinematic chain. For instance, the root segment (typically the pelvis) is firstly registered to the frame reconstruction by translating and rotating the articulated model to align it to the frame reconstruction. Shape constraints are also imposed by discarding all points in the frame reconstruction that are within proximity of the articulated model’s root segment points for the remainder of the articulated registration process. This ensures that points assigned to nonroot segments in the articulated model are not incorrectly registered to points in the root segment region of the frame reconstruction. Secondly, the child segments of the root segment are registered to the frame reconstruction using only rotation. Once again, the frame reconstruction points in proximity to the registered child segment are discarded for the remainder of the articulated registration process. If the articulated model and frame reconstruction are identical, all points belonging to the frame reconstruction should be discarded by the end of the articulated registration process. In 3D-to-3D tracking methods, texture constraints can be applied by representing each surface point in 4D (grayscale texture) or 6D (color texture) instead of just in the three dimensions corresponding to the x, y, and z coordinates. Note that the texture values must be normalized since the spatial coordinates are in different units. The time series of Pose estimates (joint rotations) is referred to as a Pose map. At times, the Pose may drift from the true solution because of a local minimum in the computed solution. One possible cause of this is that the surface of the body, at any given Pose, does not match the Visual Hull because the surface is deformable. The residual error from the articulated registration process can be used to generate a confidence score on the tracking. The initial Pose detection phase can be retriggered whenever the confidence score falls below a predefined threshold. To address this, the surface hull at one of the keyframes is used to register that frame instead of the average hull. These updated “local” hulls can be used in place of the generic hull at or near the relevant keyframes, thereby facilitating continuously enhanced Pose estimations (e.g., “resetting” the tracking process), which then continues. The nonrigid registration process can be automated by solving an optimization problem to minimize the residual error between the transformed generic model and the 3D subject representation. The articulated model can also be parameterized to facilitate the application of joint constraints in the segmental local coordinate system, or AF by Euler angles. Although prone to Gimbal lock, Tait-Bryan angles, consisting of independent rotations about the three principal axes, are also commonly used to parameterize the joint rotations because their representation is intuitive, and also because joint constraints can be directly applied. For example, joint constraints can be applied to the elbow to allow only two rotational degrees of freedom for flexion/extension and forearm pronation/supination by restricting one Tait-Bryan angle to zero rotation. Model parameterization can also include scaling parameters that facilitate adjustment of segmental geometry and thus segmental inertial properties (Fig. 6). The ability to record and analyze 3D markerless Mocap data in a live Major League Baseball game setting is an extraordinary advance in performance analytics for baseball (Fig. 6). In baseball, the important features that have been explored are the kinematics of the shoulder and elbow joints. At this point in time the level of
212
S. Cadavid and W.S. Selbie
Fig. 6 The computed Pose map for three different times in a pitch. The images display the underlying skeleton and the Visual Hull
accuracy is comparable to marker-based Mocap, partly because Marker-based Mocap does a relative poor job of tracking the arm (and shoulder joint center) compared to other segments. In other words, both methods of MoCap, Markerless and Marker-based, can struggle at times to deal with such a highly complex and ballistic movement. The benefit of a markerless approach, in this situation, is the ability to record actual in-game data, which would be impossible using markers. These markerless techniques, however, are proving to be sufficiently accurate for whole body motion and temporal analyses. This means that markerless MoCap could well be applicable in a variety of situations. Given this, the ability to record and analyze 3D markerless Mocap data for clinical assessment would be an exciting evolution in clinical movement analysis.
3D-2D Pose Estimation When only a few cameras are used to record the motion, the 3D-to-3D techniques described earlier cannot generate a sufficiently accurate surface Visual Hull from which the Stereo Reconstruction or Space Carving methods can generate accurate 3D Pose estimation. Biplanar Videoradiography (also known as Dynamic Stereo X-ray; DSX) was selected as the exemplar 3D-to-2D application for this chapter because there is evidence that abnormal mechanical joint function contributes significantly to the development and progression of many types of joint disease. Notably, joint translations of only a few millimeters are critical to estimating tissue stress or joint impingement during loaded functional movements. DSX is the only currently available technology that can achieve submillimeter bone Pose estimation accuracy during a wide variety of functional movements. The details of the DSX application are described elsewhere in this handbook, but because of its importance, it is highlighted in this chapter as well. DSX is a form of 3D-to-2D markerless MoCap that tracks and reconstructs the underlying skeletal structures of a body. The following section will give a brief
3D Dynamic Pose Estimation from Markerless Optical Data
213
Fig. 7 3D representation of a biplanar X-ray configuration when the two X-rays are synchronized. The distal femur, reconstructed from the CT data, is shown in the middle. The inline X-ray image (in line with the X-axis of the lab reference frame) is shown in the red frame; the red line is the perpendicular from the center of the X-ray image plane to the X-ray source. The offset X-ray image (offset from the X-axis of the lab reference frame) is shown in the green frame; the green line is the perpendicular from the center of the X-ray image plane to the X-ray source. For illustration, the inline X-ray image is shown after processing (smoothing and edge detection) and the offset image is shown unprocessed
overview of the techniques involved in DSX. Given a 3D representation of a bone extracted from a high-resolution CT scan of the subject, a local reference frame assigned to the bone, and a time series of X-ray images containing the bone, a Pose map is the solution of the DSX across all frames. The DSX algorithm (Bey et al. 2006, 2008; Ohnishi et al. 2010) solves for the 3D Pose by registering two noncoplanar X-ray images of a bone to two digitally reconstructed radiographs (DRRs) (Fig. 7). Given the position and orientation of an X-ray source, an X-ray image plane, and volumetric CT bone, a DRR is the projection of the CT bone onto a virtual X-ray image, using a simplified X-ray generation model (Siddon 1985; Zhao and Reader 2003). In other words, rays from the X-ray source are cast through the bone to generate a simulated X-ray with the same size and resolution as the actual X-ray. The objective function for the DSX optimization is based on a measure of the overlap between an X-ray image and a DRR. The similarity (or overlap) of the DRR of a given 3D bone Pose and an X-ray image at time stamp j is quantified as a normalized correlation (rj) for all pixels (x, y): P
I Xrayj ðx, yÞ I Xrayj I XDRRj ðx, yÞ I XDRRj r j ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2ffi I Xrayj ðx, yÞ I Xrayj I XDRRj ðx, yÞ I XDRRj xy
214
S. Cadavid and W.S. Selbie
such that IXDRR(x, y) represents the intensity of the DRR at pixel x,y; IXDRR is the mean of the DRR and IXRay(x, y) represents the intensity of the actual X-ray at pixel x,y, and IXRay ðx, yÞ is the mean of the X-ray. Registration is considered maximum similarity or minimum normalized correlation. High speed X-ray images are particularly challenging for this approach because the bone and tissue do not display uniform (or consistent) density across the images. This inconsistency means that the objective function for the registration (e.g., optimization) may identify Poses that do not appear visually to be the best solution. To compensate for the existence of local minima, this optimization problem is solved using global algorithms such as simulated annealing (Higginson et al. 2005; Ingber 2012), a Monte Carlo method in which the solution space is explored probabilistically by randomly searching near the best-known solution. It is modeled after annealing in metallurgy, in which the thermodynamic free energy of a metal decreases as its temperature decreases. In simulated annealing, as the virtual temperature cools, the algorithm searches in a smaller and smaller region around the best-known solution. Simulated annealing is an ideal optimization technique as it is not prone to finding local minima and thus a user can be confident that given a reasonable initial seed, the Pose it returns will be the optimal Pose Map across all frames. DSX has been demonstrated to have the capability to capture dynamics movements with high sampling rates (150-200 Hz) and submillimeter spatial accuracy. DSX has been used to characterize a variety of joint disorders, including changes in joint contact kinematics with knee injuries (ACL,PCL, meniscus) (Tashman et al. 2004; Gill et al. 2009; Van de Velde et al. 2009; Hoshino et al. 2013; Goyal et al. 2012; Marsh et al. 2014), dynamic aspects of patello-femoral disorders (Fernandez et al. 2008; Bey et al. 2008), femoro-acetabular impingement of the hip (Martin et al. 2011; Kapron et al. 2014), shoulder function after rotator cuff injury (Bey et al. 2011) and arthroplasty (Massimini et al. 2010), changes in intervertebral kinematics with lumbar disc degeneration (Anderst et al. 2008; Li et al. 2011), and deformation of the joint capsule and intervertebral discs with cervical spine disc fusion (Anderst et al. 2013, 2014).
Future Directions Two future directions of markerless Mocap for biomechanical applications and one future direction for DSX are highlighted; first, advances in defining subject-specific multibody segment/surface models that do not require manual identification of anatomical references frames and rigged surfaces; second, advances in physicsbased simulations that influence the markerless 3D Pose estimation and that permit the use of only one video camera; and third, 4D and hierarchical tracking of DSX data.
3D Dynamic Pose Estimation from Markerless Optical Data
215
First, impressive results achieved through Deep Learning in image classification tasks in uncontrolled environments have spawned interest in applying such techniques to human Pose estimation. Traditional classification algorithms employ hand-designed feature extraction schemes, which are naturally limited because they require substantial engineering efforts to develop, and are limited to what humans perceive as being “good” features. The advantage of Deep Learning is that Deep Learning methods employ an unsupervised approach to learning features directly from the training data. Convolutional Neural Networks (CNN), for instance, employ kernels that are convolved with an image to extract a feature representation. The kernel weights are trained at multiple resolutions to capture features at different scales. Toshev and Szegedy (2014) proposed a method consisting of a cascaded set of CNNs to estimate Pose from a single 2D view in uncontrolled environments. A vast dataset of normalized images and corresponding Poses, spanning a variety of activities, is used to train each predictor. The initial predictor estimates an initial Pose for a subject from a single viewpoint, which aims to obtain the same output as the initial Pose detection method described in the previous section. Additional Pose predictors are then trained for each body segment in order to refine the localization of the segments. The authors report state-of-the-art or better performance on four academic benchmarks including the Frames Labeled In Cinema dataset, the Leeds Sports Dataset, the Buffy dataset, and the Image Parse dataset. Future directions in Markerless Mocap include the application of Deep Learning to point clouds to estimate a 3D Pose (Wu et al. 2015), incorporating multiple calibrated views of a Pose to train a Deep Learning classifier, and validating these methods against the HumanEva II dataset (Sigal et al. 2010) and Faust dataset (Bogo et al. 2014), which consists of marker based ground-truth data corresponding to motion trial video data. Second, for the markerless Mocap solution, at each video frame, the estimated Pose is a kinematic solution based on registering the vertices of the Visual Hull to the multibody model. The solution, however, is not guaranteed to be dynamically consistent (i.e., to satisfy the laws of mechanics). Physics-based models can be used to drive the Pose estimation from the markerless Mocap system toward dynamic consistency in a controlled way through a Bayesian approach where the markerless Pose is fused statistically with a physical simulation. Motivated by the abundance of conventional video footage, recent approaches based on Deep Learning are aiming to elaborate the physics-based simulations to estimate 3D Pose from a single optical camera (Toshev and Szegedy 2014; Wandt et al. 2016; Wang et al. 2014). Third, for the DSX application, recent research on 4D and hierarchical tracking is promising (Tashman et al. 2017). In a 4D solution, the Pose is not solved at each frame, but rather solved across all frames of data simultaneously by treating the Pose map as a spline across time rather than a collection of discrete Poses. One of the advantages of this approach is that the X-ray emitters can be run slightly out of phase, which dramatically reduces the X-ray scatter in each image.
216
S. Cadavid and W.S. Selbie
Cross-References ▶ 3D Dynamic Pose Estimation from Marker-Based Optical Data ▶ 3D Dynamic Probabilistic Pose Estimation from Data Collected Using Cameras and Reflective Markers ▶ Estimation of the Body Segment Inertial Parameters for the Rigid Body Biomechanical Models Used in Motion Analysis ▶ Measurement of 3D Dynamic Joint Motion Using Biplane Videoradiography ▶ Physics-Based Models for Human Gait Analysis ▶ Three-Dimensional Reconstruction of the Human Skeleton in Motion
References Anderst WJ, Vaidya R, Tashman S (2008) A technique to measure three-dimensional in vivo rotation of fused and adjacent lumbar vertebrae. Spine J 8:991–997 Anderst WJ, Donaldson WF, Lee JY, Kang JD (2014) In vivo cervical facet joint capsule deformation during flexion-extension. Spine J 39(8):514–520 Anderst W, Donaldson W, Lee J, Kang J (2013) Cervical disc deformation during flexionextension in asymptomatic controls and single-level arthrodesis patients. J Orthop Res 31 (12):1881–1889 Anguelov D, Srinivasan P, Koller D, Thrun S, Rodgers J, Davis J, (2005) SCAPE: shape completion and animation of people. In: ACM transactions on graphics (TOG), vol 24, no 3. ACM, pp 408–416 Balan AO, Sigal L, Black MJ, Davis JE, Haussecker HW (2007) Detailed human shape and pose from images. In: 2007 IEEE conference on computer vision and pattern recognition. IEEE, pp 1–8 Bay H, Tuytelaars T, Van Gool L (2006) Surf: speeded up robust features. In: European conference on computer vision. Springer, Berlin/Heidelberg, pp 404–417 Besl PJ, McKay ND (1992) Method for registration of 3-D shapes. In: Robotics-DL tentative. International Society for Optics and Photonics, pp 586–606 Bey MJ, Zauel R, Brock SK, Tashman S (2006) Validation of a new model-based tracking technique for measuring three-dimensional, in vivo glenohumeral joint kinematics. J Biomech Eng 128:604–609 Bey MJ, Kline SK, Tashman S, Zauel R (2008) Accuracy of biplane X-ray imaging combined with model-based tracking for measuring in-vivo patellofemoral joint motion. J Orthop Surg Res 3:38 Bey MJ, Peltz CD, Ciarelli K, Kline SK, Divine GW, van Holsbeeck M, Muh S, Kolowich P, Lock T, Moutzouros V (2011) In vivo shoulder function after surgical repair of a torn rotator cuff: Glenohumeral joint mechanics, shoulder strength, clinical outcomes, and their interaction. Am J Sports Med 10:2117–2129 Bogo F, Romero J, Loper M, Black MJ (2014) FAUST: Dataset and evaluation for 3D mesh registration In: Proceedings IEEE Conference on computer vision and pattern recognition (CVPR), pp 3794–3801 Bogo F, Black MJ, Loper M, Romero J (2015) Detailed full-body reconstructions of moving people from monocular RGB-D sequences. In: International conference on computer vision (ICCV), pp 2300–2308 Brainerd EL, Baier DB, Gatesy SM, Hedrick TL, Metzger KA, Crisco JJ (2010) X-ray reconstruction of moving morphology (XROMM): precision, accuracy and applications in comparative biomechanics research. J Exp Zool 313A:262–279
3D Dynamic Pose Estimation from Markerless Optical Data
217
Cappozzo A, Catani F, Leardini A, Benedetti MG, Della Croce U (1996) Position and orientation in space of bones during movement: experimental artefacts. Clin Biomech 11(2):90–100 Cheung KMG, Baker S, Kanade T (2003) Shape-from-silhouette of articulated objects and its use for human body kinematics estimation and motion capture. In: Computer vision and pattern recognition, 2003. Proceedings. 2003 I.E. computer society conference on, vol 1, IEEE, pp 1–77 Corazza S, Mündermann L, Gambaretto E, Ferrigno G, Andriacchi TP (2010) Markerless motion capture through visual hull, articulated icp and subject specific model generation. Int J Comput Vis 87(1–2):156–169 Della Croce U, Leardini A, Chiari L, Cappozzo A (2005) Human movement analysis using stereophotogrammetry. Part 4: assessment of anatomical landmark misplacement and its effects on joint kinematics. Gait Posture 21(2):226–237 Fernandez JW, Akbarshahi M, Kim HJ, Pandy MG (2008) Integrating modelling, motion capture and x-ray fluoroscopy to investigate patellofemoral function during dynamic activity. Comput Methods Biomech Biomed Engin 11(1):41–53 Gill TJ, Van de Velde SK, Wing DW, Oh LS, Hosseini A, Li G (2009) Tibiofemoral and Patellofemoral kinematics following reconstruction of an isolated posterior cruciate ligament injury: in vivo analysis during lunge. Am J Sports Med 37(12):2388–2385 Giphart JE, Zirker C, Myers C, Pennington WW, LaPrade RF (2012) Accuracy of a contour-based biplane fluoroscopy technique for tracking knee joint kinematics of different speeds. J Biomech 45:2935–2938 Gower JC (1975) Generalized procrustes analysis. Psychometrika 40(1):33–51 Goyal K, Tashman S, Wang JH, Li K, Zhang X, Harner C (2012) In vivo analysis of the isolated posterior cruciate ligament-deficient knee during functional activities. Am J Sports Med 40(4):777–785 Haque MA, Anderst W, Tashman S, Mari GE (2013) Hierarchical model-based tracking of cervical vertebrae from dynamic biplane radiographs. Med Eng Phys 35(7):994–1004 Harris C, Stephens M (1988) A combined corner and edge detector. In: Alvey vision conference, vol 15, p 50 Higginson JS, Neptune RR, Anderson FC (2005) Simulated parallel annealing within a neighborhood for optimization of biomechanical systems. J Biomech 38:1938–1942 Hoshino Y, Fu FH, Irrgang JJ, Tashman S (2013) Can joint contact dynamics be restored by anterior cruciate ligament reconstruction? Clin Orthop Relat Res 471(9):2924–2931 Ingber L (2012) Adaptive simulated annealing. In: Oliveira H, Petraglia A, Ingber L, Machado M, Petraglia M (eds) Stochastic global optimization and its applications with fuzzy adaptive simulated annealing. Springer, New York, pp 33–61 KaewTraKulPong P, Bowden R (2002) An improved adaptive background mixture model for realtime tracking with shadow detection. In: Video-based surveillance systems. Springer, New York, pp 135–144 Kapron AL, Aoki SK, Peters CL, Maas SA, Bey MJ, Zauel R, Andersen A (2014) Accuracy and feasibility of dual fluoroscopy and model-based tracking to quantify in vivo hip kinematics during clinical exams. J Appl Biomech 30(3):461–470 Keskin C, Kıraç F, Kara YE, Akarun L (2013) Real time hand pose estimation using depth sensors. In: Consumer depth cameras for computer vision. Springer, London, pp 119–137 Kutulakos KN, Seitz SM (2000) A theory of shape by space carving. Int J Comput Vis 38(3):199–218 Li W, Wang S, Xia Q, Passias P, Kozanek M, Wood K (2011) Lumbar facet joint motion in patients with degenerative disc disease at affected and adjacent levels: an in vivo biomechanical study. Spine 36(10):629–637 Loper M, Mahmood N, Black MJ (2014) MoSh: motion and shape capture from sparse markers. ACM Trans Graph 33(6):220:1–220:13 Lorensen WE, Cline HE (1987) Marching cubes: a high resolution 3D surface construction algorithm. In: ACM siggraph computer graphics, vol 21, no 4. ACM, pp 163–169
218
S. Cadavid and W.S. Selbie
Lourakis MI, Argyros AA (2009) SBA: a software package for generic sparse bundle adjustment. ACM Trans Mathemat Software 36(1):2 Lowe DG (1999) Object recognition from local scale-invariant features. In: Computer vision, 1999. The proceedings of the seventh IEEE international conference on, vol 2, IEEE, pp 1150–1157 Marsh C, Martin DE, Harner C, Tashman S (2014) Effect of posterior horn medial meniscus root tear on in vivo knee kinematics. Orthop J Sports Med 2(7):1–7 Martin DE, Greco NJ, Klatt BA, Wright VJ, Anderst WJ, Tashman S (2011) Model-based tracking of the hip: implications for novel analyses of hip pathology. J Arthroplast 26(1):88–97 Massimini DF, Li G, Warner JP (2010) Glenohumeral contact kinematics in patients after total shoulder arthroplasty. J Bone Joint Surg Am 92(4):916–926 Nakamura T, Matsumoto J, Nishimaru H, Bretas RV, Takamura Y, Hori E, Ono T, Nishijo H (2016) A markerless 3D computerized motion capture system incorporating a skeleton model for monkeys. PLoS One 11(11):e0166154 Ohnishi T, Suzuki M, Nawata A, Naomoto S, Iwasaki T, Haneishi H (2010) Three-dimensional motion study of femur, tibia, and patella at the knee joint from bi-plane fluoroscopy and CT images. Radiol Phys Technol 3:151–158 Piccardi M (2004) Background subtraction techniques: a review. In: Systems, man and cybernetics, 2004 I.E. international conference on, vol 4, IEEE, pp 3099–3104 Marcard T, Pons-Moll G, Rosenhahn B (2016) Human pose estimation efrom video and IMUs. Trans Patt Anal Mach Intellig 38:1533–1547 Pons-Moll G, Romero J, Mahmood N, Black M (2015) Dyna: a model of dynamic human shape in motion. ACM Trans Graph 34(4):120:1–120:14 Rosten E, Drummond T (2006, May) Machine learning for high-speed corner detection. In: European conference on computer vision. Springer, Berlin/Heidelberg, pp 430–443 Rublee E, Rabaud V, Konolige K, Bradski G (2011) ORB: An efficient alternative to SIFT or SURF. In: 2011 international conference on computer vision, IEEE, pp 2564–2571 Seitz, S.M., Curless, B., Diebel, J., Scharstein, D. and Szeliski, R., 2006, June. A comparison and evaluation of multi-view stereo reconstruction algorithms. In: 2006 I.E. computer society conference on computer vision and pattern recognition (CVPR’06), vol 1, IEEE, pp 519–528 Shotton J, Sharp T, Kipman A, Fitzgibbon A, Finocchio M, Blake A, Cook M, Moore R (2013) Real-time human pose recognition in parts from single depth images. Commun ACM 56(1):116–124 Siddon RL (1985) Fast calculation of the exact radiological path for a three-dimensional CT array. Med Phy 12:252–255 Sigal L, Balan AO, Black MJ (2010) Humaneva: synchronized video and motion capture dataset and baseline algorithm for evaluation of articulated human motion. Int J Comput Vis 87(1–2):4–27 Stoll C, Hasler N, Gall J, Seidel HP, Theobalt C (2011) Fast articulated motion tracking using a sums of gaussians body model. In: 2011 international conference on computer vision, IEEE, pp 951–958 Tashman S, Collon D, Anderson K, Kolowich P, Anderst W (2004) Abnormal rotational knee motion during running after anterior cruciate ligament reconstruction. Am J Sports Med 32(4):975–983 Tashman S, Princehorn J, Penatto S, Andherst W (2017) Intelligent algorithms for tracking threedimensional skeletal movement from radiographic image sequences. US patent # 9538940 B2 Toshev A, Szegedy C (2014) Deeppose: human pose estimation via deep neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1653–1660 Triggs B, McLauchlan PF, Hartley RI, Fitzgibbon AW (1999) Bundle adjustment – a modern synthesis. In: International workshop on vision algorithms. Springer, Berlin/Heidelberg, pp 298–372 Van de Velde SK, Gill TJ, Li G (2009) Evaluation of kinematics of anterior cruciate ligamentdeficient knees with use of advanced imaging techniques, three-dimensional modeling techniques, and robotics. J Bone Joint Surg Am 91(Suppl 1):108–114
3D Dynamic Pose Estimation from Markerless Optical Data
219
Wandt B, Ackermann H, Rosenhahn B (2016) 3d reconstruction of human motion from monocular image sequences. Trans Pattern Analy Mach Intellig 38(8):1505–1516 Wang C, Wang Y, Lin Z, Yuille A, Gao W (2014). Robust estimation of 3d human poses from a single image. In: Conference on computer vision and pattern recognition (CVPR) Weiss A, Hirshberg D, Blanc MJ (2011) Home 3D body scans from noisy image and range data. In: ICCV ’11 proceedings of the 2011 international conference on computer vision, pp 1951–1958 Wu Z, Song S, Khosla A, Yu F, Zhang L, Tang X, Xiao J (2015) 3d shapenets: A deep representation for volumetric shapes. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1912–1920 Zhao H, Reader AJ (2003). Fast ray-tracing technique to calculate line integral paths in voxel arrays. In: Proceedings of the IEEE nuclear science symposium and medical imaging conference, pp 2808–2812 Zivkovic Z (2004) Improved adaptive Gaussian mixture model for background subtraction. In: Pattern recognition, 2004. ICPR 2004. Proceedings of the 17th international conference on, vol 2, IEEE, pp 28–31
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial Measurement Units Andrea Cereatti, Ugo Della Croce, and Angelo M. Sabatini
Abstract
This chapter deals with the estimation of human kinematics using magneto and inertial sensing technology. A magneto-inertial measurement unit typically embeds a triaxial gyroscope, a triaxial accelerometer, and a triaxial magnetic sensor in the same assembly. By combining the information provided by each sensor within a sensor fusion framework, it is possible to determine the unit orientation with respect to a common global coordinate system. Recent advances in the construction of microelectromechanical system devices have made possible the manufacturing of small and light devices. These advances have widened the range of possible applications to include areas such as human movement. This chapter aims at providing the reader with a picture of the state of the art in the measurement and estimation methods for the description of human joint kinematics using magneto-inertial sensing technology. In the first section, fundamental concepts of rigid body kinematics are introduced with special reference to magneto-inertial measurements. Then a short description of the operational A. Cereatti (*) Department POLCOMING, University of Sassari, Sassari, Italy Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, University of Sassari, Sassari, Italy Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy e-mail: [email protected] U. Della Croce Department POLCOMING, University of Sassari, Sassari, Italy Interuniversity Centre of Bioengineering of the Human Neuromusculoskeletal System, University of Sassari, Sassari, Italy e-mail: [email protected] A.M. Sabatini The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_162
221
222
A. Cereatti et al.
characteristics of accelerometers, gyroscopes, and magnetometers is provided. The third section reports theory and methods for the estimation of the orientation and position of magneto-inertial measurement units along with the implementation of a Kalman filter for 3D orientation estimate as an example. In the last section, a critical review of the most common methodologies for the joint kinematic estimation is reported. Keywords
Joint mechanics • Acceleration • Angular velocity • Orientation • Position • Multisegmental model • Multibody • Anatomical coordinate system • Joint kinematics • Wearable sensors • Kalman filter • Pose Abbreviations
ALI ARW ACS BCS CoR CS DoFs EKF FUN KF GCS IMU MCS MEMS (M)IMU MUL NEMS VRW h, i [q]
Anatomical landmark identification Angle Random Walk Anatomical coordinate system Body-fixed coordinate system Center of rotation Coordinate system Degree of freedom Extended Kalman filter Functional Kalman filter Global coordinate system Inertial measurement unit MIMU coordinate system Microelectromechanical systems (Magneto)-inertial measurement unit Manual Unit Alignment Nano-electromechanical systems Velocity Random Walk Dot product between vectors Quaternion multiplication Skew-symmetric operator
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rigid Body Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magneto-Inertial Measurement Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Orientation and Position Estimates Using MIMU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three-Dimensional Human Joint Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joint Positional Kinematic Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
223 224 224 228 230 233 240 241 242 242
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial. . .
223
Introduction Human movement kinematics requires the description of the displacements, velocities, and accelerations, with respect to a global coordinate system (GCS), of every bony segment modeling the portion of skeletal system under analysis. In general, the body segments are assumed to be perfectly rigid and therefore to constitute a single rigid body with the underlying bone. Adjacent body segments are constrained through ideal joint models. Since the majority of the human joints allow only small relative linear displacements, whose amplitude is comparable to the errors associated to their estimate, only joint angular displacements are generally considered. This chapter deals with the estimation of the human kinematics using magneto and inertial sensing technology. The term inertial navigation refers to a set of techniques that exploit measurements of linear accelerations and angular velocities, with the aim of estimating the position and orientation (pose) of an object in the three-dimensional (3D) space relative to a known starting point, orientation, and velocity (Titterton and Weston 2004). An inertial measurement unit (IMU) is commonly used to measure linear accelerations and angular velocities. Typically, an IMU embeds a triaxial gyroscope and a triaxial accelerometer; another type of sensor that is commonly integrated in an IMU is a triaxial magnetic sensor, which measures the strength and direction of the local magnetic field, allowing the north direction to be found (Barbour and Schmidt 2001). Triaxial means that the sensor sensitivity axes are three, and they are mutually orthogonal to span the whole space and define the IMU-based coordinate system. The term MIMU (magneto-inertial measurement unit) is popularly used to denote a device that integrates accelerometer, gyroscope, and magnetic sensor in the same assembly (Bergamini et al. 2014). Inertial navigation was first developed for applications including navigation of aircraft, tactical and strategic missiles, submarines, and ships. Recent advances in the construction of microelectromechanical system (MEMS) devices have made possible the manufacturing of small and light IMUs. These advances have widened the range of possible applications to include areas such as human motion. Several key factors are behind the success of these sensing technologies. First, (M)IMUs are self-contained, in the sense that they do not rely on any external infrastructure to be operational. Second, since inertial and magnetic sensors are heavily used in the consumer electronics market, their price keeps dropping, while their performance improves. Lastly, the move from wearable measurement systems to pervasive systems made possible by the MEMS/NEMS (nano-electromechanical systems: NEMS) technology opens up new perspectives for motor performance assessment and monitoring. This chapter aims at providing the reader with a picture of the state of the art in the measurement and estimation methods for the description of the human joint kinematics using (M)IMUs. In the first section of the chapter, fundamental concepts of rigid body kinematics are introduced with special reference to magneto-inertial measurements. Then a short description of the operational characteristics of accelerometers, gyroscopes, and magnetometers is provided. The third section reports
224
A. Cereatti et al.
theory and methods for the estimation of the orientation and position of (M)IMUs along with the implementation of a Kalman filter for 3D orientation estimate as an example. In the final section, a critical review of the most common methodologies for joint kinematics estimation is reported.
State of the Art Rigid Body Kinematics Let B be a rigid body and P a point of B (Fig. 1). Two coordinate systems are introduced: the global (earth-fixed) coordinate system (GCS), specified by the origin OG and the right-handed orthonormal basis G = {Gx Gy Gz}, and the body-fixed coordinate system (BCS), specified by the origin OB and the right-handed orthonormal basis B = {Bx By Bz}. The motion of the body B is described by the translation of the origin OB and the rotation of ! BCS with respect to GCS. The translation is given by the vector G b ¼ OG OB of the ! origin OB relative to GCS. The vector G p ¼ OG P gives the coordinates of P relative ! to GCS; finally, the vector G r ¼ OB P is the vector from the origin OB to P (as seen from GCS): G
p ¼ G b þ G r,
(1)
With explicit notation, all vectors in Eq. 1 are measured in GCS. The vector Bω represents the angular velocity, resolved in the BCS, and it describes the rotational speed of B and its axis of rotation (Fig. 1). Since the body is rigid, the magnitude of G r is constant during the body motion. The vector Gr can be expressed in terms of the time-independent vector Br from OB to P as seen in BCS:
Fig. 1 Rigid body motion
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial. . . B
r ¼ BG CG r,
225
(2)
where C ¼ BG C is called the rotation matrix from GCS to BCS (Shuster 1993). An alternative representation is provided by the quaternion q ¼ BG q. The quaternion q is defined by: q ¼ ½q q4 T :
(3)
It is composed of the scalar component q4 and the vector component given by q = [qx qy qz]T. For the quaternion to be a valid parameterization of rotation, the following normalization constraint must be enforced: jqj2 þ q24 ¼ q2x þ q2y þ q2z þ q24 ¼ 1:
(4)
The quaternion multiplication between two generic quaternions q and h is defined as: qh¼
q4 h þ h4 q q h : q 4 h 4 qT h
(5)
A generic vector Gp can be transformed from one coordinate system (e.g., GCS) to another (e.g., BCS), by pre- and post-multiplying its quaternion (with scalar part equal to zero) by the rotation quaternion BG q and its inverse (i.e., the same quaternion with the vector part changed by sign), respectively: B
p ¼ BG q
G p BG q1 0
(6)
According to the Euler’s theorem, the most general motion of a rigid body with one point fixed is a rotation by an angle θ (rotation angle) about some axis n (rotation axis). This yields another representation of the rigid body orientation in terms of a rotation vector: θ ¼ θn:
(7)
The rotation vector is related to the quaternion as follows:
q ¼ qx
qy
qz
T
θ ¼ sin n, 2
θ q4 ¼ cos 2
(8)
The rotation matrix C can be expressed as a function of the quaternion q (Shuster 1993)
C ¼ 2q24 1 I33 ½q 2q4 þ qqT 2,
(9)
226
A. Cereatti et al.
where In n is the n n identity matrix (n = 3) and the skew-symmetric operator: 2
0 ½q ¼ 4 qz qy
qZ 0 qx
3 qy qx 5, 0
(10)
and denotes the matrix notation for the cross product. In other words, the skewsymmetric matrix can be used to represent the cross product between the vector q and a generic vector t in terms of the matrix-vector multiplication [q]t. When the BCS is moving with respect to the GCS, the rotation matrix C ¼ G BC can be shown to be the solution to the matrix differential equation: C_ ¼ ½ω C,
(11)
where ω = Bω. A triaxial gyroscope with the sensitivity axes aligned along the directions of BCS is customarily used in (M)IMUs to provide the (noisy) measurements of the angular velocity vector. Equation 11 can be reformulated in an equivalent form involving the quaternion q¼G B q and its first-order time derivative: 1 1 ½ω q_ ¼ ΩðωÞq ¼ 2 2 ωT
ω q: 0
(12)
The solution to this system of first-order linear differential equations from known initial conditions provides therefore the orientation of the rigid body relative to the GCS. Furthermore, time differentiation of (11) yields: € ¼ ½ω _ þ ½ω2 C: C
(13)
Using Eqs. 1 and 2, the velocity and the acceleration of the point P (relative to GCS) can then be written: G_ G _B p ¼ G b_ þ G B C r ¼ b þ ½ω r G€ € B r ¼ Gb € þ GC € þ ½ω _ þ ½ω2 G r, p ¼ Gb B G_
A triaxial accelerometer located in P senses, resolved in BCS, the so-called € B g , where Bg is the constant-orientation gravity vecspecific force B a ¼ B p tor (Ligorio and Sabatini 2016); for example, when the basis vector Gz is aligned parallel to the gravity vector, Gg = [0 0 g]T, with g = 9.81 m/s2. Let u = Br and θi , i = x , y , z denote the location and sensing directions of the accelerometer with respect to BCS, which are all time independent. Using (14), the components of the
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial. . .
227
specific force (accelerometer output) a = [ax ay a z]Tcan also be expressed with all quantities resolved in the BCS as: ai ðu, θi Þ ¼
D
E _ þ ½ω2 u, θi , i ¼ 1, 2, 3 b B g þ ½ω
B€
(15)
where h, i denotes the dot product between vectors. The first term within the dot product expresses the specific force as the additive combination of linear acceleration, gravity vector, and angular acceleration; the angular acceleration includes the _ tangential acceleration ½ωu and the centripetal acceleration [ω][ω]u. Suppose that the location of the accelerometer is where the origin of the BCS lies, i.e., u = 0. The angular acceleration is therefore null; without loss of generality, we can also assume that the
sensitive axes of the triaxial accelerometer are oriented along the directions of Bx By Bz , namely, θ1 = [1 0 0]T , θ2 = [0 1 0]T , θ3 = [0 0 1]T. As for tracking the origin of the BCS, it is necessary, first, to rotate the (noisy) measured specific force from the BCS to the GCS using the rotational matrix computed from the integration of Eq. 11, or equivalently Eq. 12; second, the gravity vector contribution must be canceled by adding the known expression of Gg to the rotated measured specific force: G
G B BC a
€¼ p
þ G g:
(16)
When available in the GCS, the linear acceleration can be integrated once to obtain velocity and again to obtain displacement (strap-down approach to inertial navigation).
Determination of the Center of Rotation of a Rigid Body Let us consider the rigid body B constrained to a rigid frame through a spherical joint; the body can then only experience a pure rotational motion around the center of rotation (CoR). According to Eq. 14, assuming OBOG and coinciding with the CoR, the acceleration of a point P can be expressed as: G
€ ¼ ½ω _ þ ½ω2 G r, p
(17)
After some algebraic manipulation, Eq. 2 can be rearranged as: €, KG r ¼ G p
(18)
where 2
ω2y ω2z
6
K¼6 4 ω_ z þ ωx ωy
ωx ωz ω_ y
ωx ωy ω_ z 2
ωx ω2z
ω_ x þ ωy ωz
ω_ y þ ωx ωz
3
7 7 ωy ωz ω_ x 5 ω2x ω2y
(19)
228
A. Cereatti et al.
Equation 18 is linear in the unknown vector Gr, which represents the position € can be obtained from the accelerometric vector of CoR in the GCS. The vector G p measurements according to (16), and ω can be obtained from the gyroscope readings. The position vector of CoR can be then expressed in the BCS using either the rotation matrix or the rotation quaternion from GCS to BCS. In the presence of noise, a more reliable estimate of Gr can be obtained by computing (18), for each of the N sampled instants of time, recorded during a pure rotational motion of the rigid body, to obtain an over determined linear system which can be solved using a least-square technique.
Determination of the Axis of Rotation of a Rigid Body Let us consider the rigid body B constrained to a rigid frame through a revolute joint, which is rotating with an angular velocity ω around the single axis of rotation n. The most straightforward solution to compute the direction of n is from its angular velocity: n¼
ω , kωk
(20)
Alternatively, n can be also obtained from Eq. 8: n¼
q , θ ¼ 2cos1 ðq4 Þ, θ sin 2
(21)
In the presence of noise, a more reliable estimate of r can be estimated either by selecting only angular velocity above a given threshold or by averaging the quaternion over the N observations (Prentice 1986).
Magneto-Inertial Measurement Technology Measurement System Description (M)IMUs for applications in the areas of human motion fall in the category of so-called strap-down systems. Since inertial sensors are rigidly mounted on the device, output quantities are measured in the BCS rather than the GCS (Titterton and Weston 2004). To track the (M)IMU orientation, the signals from the gyroscopes are time integrated. To track position, the signals from the accelerometer must be resolved into earth-fixed coordinate system (GCS) using the computed orientation and then integrated from known initial conditions. This procedure is shown in Fig. 2. In strap-down IMUs, the signals produced by the inertial sensors are resolved mathematically, prior to the calculation of navigational information. This reduces the mechanical complexity of the inertial navigation system, as it is implemented in the classical applications of inertial navigation technology, i.e., stable platform technique, thus decreasing the cost and size of the system and consequently increasing its
Three-Dimensional Human Kinematic Estimation Using Magneto-Inertial. . .
229
Fig. 2 Strap-down approach to inertial navigation
reliability. Nowadays, the processing speed and low cost of modern computers and microcomputers allow for the implementation of wearable strap-down IMUs for applications in human motion.
Measurement Characteristics of State-of-the-Art MEMS It is commonplace to distinguish different categories, or grades, of inertial sensors, which group them according to their expected performance, namely, in decreasing order of performance, “marine and navigation,” “tactical,” “industrial,” and “automotive and consumer” grades (Yazdi et al. 1998). Table 1 depicts the expected performance in terms of parts per million (ppm) of scale-factor stability (i.e., how well the sensor reproduces the sensed angular velocity or acceleration) and /h or m/s2/h of inherent bias stability (i.e., the error independent of angular velocity or acceleration). While these performance factors are not the only ones that influence sensor selection, they are useful for comparison purposes. Quite invariably, the MEMS technologies that, for reasons of cost, complexity, size, and weight, are compatible with the requirements of human motion studies fall within the “automotive and consumer” grade. It is noted that in the absence of rotation (acceleration), the gyroscope (accelerometer) output is the sum of white noise and a slowly varying function (bias). The parameters Angle Random Walk (ARW) and Velocity Random Walk (VRW) reported in Table 1 are customarily used to quantify the white noise strength (in alternative to the RMS per square root of measurement bandwidth). These noise specifications describe the average deviation occurring when signals from the gyroscope and the accelerometer are integrated and when their actual estimation is based on Allan variance computation (El-Sheimy et al. 2008). Although the bias drift could be defined in different ways, the values reported in Table 1 are intended as the peak-to-peak value of the bias (Yazdi et al. 1998). The accuracy specifications of inertial sensors reported in Table 1 cannot be directly translated into pose error estimates. Qualitatively, longer time horizons and higher accuracy are achievable when higher-grade inertial sensors are used in a given application. However, other problems exist, which can be particularly critical because of the low performance requirements of MEMS/NEMS technologies. First, the difficulty of correctly interpreting the acceleration signals, when the component due to the gravity field (vertical reference) coexists with the component related to the
230
A. Cereatti et al.
Table 1 IMU accuracy specifications
Gyroscopes Angle random walk [ /h/√Hz] Bias drift [ /h] Scale factor stability [ppm] Accelerometers Velocity random walk [g/√Hz] Bias drift [m/s2/h] Scale factor stability [ppm]
Automotive and consumer
Tactical
Marine and navigation
>1 >100 >2000
0.1–0.5 5–50 500–1500
5000
50–75 500–1,000 1000–3000
100 steps recorded in a 24-h period. They also noted that only 75% of their sample was able to capture 5 days of monitoring. These results may be influenced by the attachment of the SW with a strap versus a knit cuff, as has been employed in previous work (Bjornson et al. 2007, 2014).The definition of the metric of “day” should be clearly defined and consistent in the processing of raw SW data with the proprietary software, to allow accurate interpretation pre-/post-intervention and across studies. The construct validity of SW watch monitoring in day-to-day life was examined by comparison to gait-lab-based walking speed, a summary score of gait deviation (Gait Deviation Index, GDI) and distance walked in 6 min (Wilson et al. 2015). Examining a cohort of 55 youth with CP, primarily at GMFCS level I, the authors documented a moderate relationship between GDI and SW average strides/day (r = 0.58) and that as the strides/day increase so does GDI. SW stride activity was also significantly correlated to lab-based walking speed and walking distance. This work suggests that interventions that improve gait kinematics (orthopedic surgery, orthotics) or decrease overall gait deviations (increase in GDI score toward normal) may enhance community walking activity as well. Replication of this work across lower functioning ambulatory children and pre-/post-intervention is needed to further understand this potential relationship. StepWatch capture of habitual walking activity has been employed as an outcome within a randomized intervention testing the effect of a 6-month physical activity stimulation program in the Netherlands (van Wely et al. 2014). No significant difference was documented in walking level or intensity between intervention groups at 6 and 12 months post. Habitual walking activity with and without current AFO prescription was examined with daily walking activity and intensity with the SW in 2016 (Bjornson et al. 2016). This work documented no group level condition effect (AFO ON vs. OFF) for community walking levels for clinically prescribed AFOs in 11 children with diplegia, across GMFCS levels I–III and various barefoot walking patterns (equinus, jump gait, and crouch gait). Two participants who did exhibit improved walking activity levels were prescribed similar AFO prescriptions. These studies suggest that monitoring with the SW is feasible and has potential to provide “real-world” information to inform clinical care and research outcomes.
Physical Activity in Children with CP (Actigraph) A 2012 study utilizing the ActiGraph GT1M examined 23 ambulatory and non-ambulatory adolescents with CP (13.5, SD 2.6 year) at GMFCS levels I–IV over a 7-day period. Participants wore the device from 540.5 to 859.2 min per day and engaged in 89.5 47.1 min of LPA, 17.8 16.9 min of MPA, 12.0 14.4 min of VPA, and 30.7 30.3 min of MVPA. Youth classified at GMFCS level IV presented with lower levels of LPA, MPA, and MVPA compared with level I (P < 0.05). Similarly, youth at level III demonstrated lower levels of
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1023
MPA and MVPA compared with level I. No differences were seen between levels I and II for any intensity. MVPA and GMFCS levels were negatively correlated (minutes/day: τ = 0.65, P < 0.001) (Gorter et al. 2012). A 2015 study of 102 children (11 2 years) with spastic hemiplegia classified at GMFCS levels I and II found that only 25% of participated in 60 min of MVPA on at least one of 4 days. The mean totals for the group were 438 counts/min and 7,541 steps per day and spent 8:36 h (72% of recorded time) in sedentary time, 2:38 h in light activities (22% of recorded time), and 0:44 h (6% of recorded time) in MVPA. There were no significant differences in physical activity between children classified at GMFCS levels I and II. Children were significantly more physically active than adolescents. Boys were significantly more physically active than girls. More steps were recorded on weekdays than on weekend days (Mitchell et al. 2015).
Physical Activity in Children with CP Uptimer The Uptimer is a device validated to measure time spent upright and was used in a study of 300 children and youth with CP. The authors found that compared to 5.6 h per day of “uptime” in able-bodied peers, youth with CP described to have hemiplegia (n = 115) spent 5.1 h/day in upright time, youth with diplegia (n = 113) spent 2.5 h/day in upright time, and youth with quadriplegia (n = 72) spent 0.5 h/day in upright time. Analysis of variance revealed significant differences between all groups (P < 0.001), including the nondisabled comparison group (Pirpiris and Graham 2004). Walking Accelerometry Combined with Global Positioning System (GPS) in Children with CP A synchronization of SW walking activity and Global Positioning System (GPS) data has been piloted in 12 ambulatory children with CP (R21 HD 077186) who underwent 20 sessions of short burst interval treadmill training (SBLTT) (Bjornson, Moreau, Hurvitz, Kerfeld 2016 unpublished data). SW accelerometry walking activity data were time matched with GPS records to document walking in the home and community over a 7-day sample through distance walked, average strides per day, and percent of time each day spent in low (1–30 strides/min), moderate (31–60 stride/min), and high (>60 stride/min) stride rates. Combined StepWatch/GPS pilot data demonstrated an increase in the percentage of overall strides per day ambulated in the community setting from 44.2% to 49.8%, suggesting an increase in community walking participation at 6 weeks post-SBLTT training (Fig. 2). Figure 3 illustrates a StepWatch/GPS a synchronization map for a child with CP at GMFCS level I over one measurement day of walking activity in the Seattle, Washington, area. The StepWatch data were classified into intensity levels based on stride rate. Each GPS location was matched with the walking intensity level at the same time point, allowing measurement of where walking across these levels occurred in a spatial context. For this child, the moderate stride activity occurred at school and a soccer field. This preliminary data provides feasibility of this novel combined SW/GPS methodology and potential sensitivity for the amount and location of community strides and intensity levels.
1024
K.F. Bjornson and N. Lennon
Fig. 2 Percentage of total walking (avg. strides/day) in the community (blue) vs in the home (red) after 20 sessions of interval treadmill training (n = 12)
Fig. 3 Synchronized StepWatch/GPS map in Seattle for a child with CP (GMFCS I). X = no walking (nw); Ο = low stride rates; ☐ = moderate stride rates; Δ = home
Summary Our knowledge of the real-world physical activity and walking activity habits of children and youth with CP has expanded over the past decade. The StepWatch, the Actigraph, and Uptimer have provided data to show that ambulatory youth with CP, those at GMFCS levels I, II, and III, are more sedentary, spend less time upright, spend less time being physically active, and take fewer steps per day than TDY. Studies using these devices have shown that time upright, time active, and amount of walking are strongly associated with motor ability, as reported by GMFCS classification levels. Walking and activity habits have similarities to TDY in that school-age
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1025
children with CP take more steps and have higher PA levels than adolescents with CP. The PA and walking activity habits of school-age children have also revealed that youth with CP tend to take more steps on school days compared to non-school days. This recent addition of knowledge about the activity habits of children and youth with CP can be utilized on a broad scale to tailor PA opportunities and programs to promote greater activity and participation and can be used on an individual level in physical therapy and home programming. We did not identify for this review studies utilizing the BMSW or the VitaMove monitors to report on PA and WA for children and youth with CP, although we acknowledge that this is not an exhaustive review.
Monitoring of Walking and Physical Activity: Clinical Implications for Children CP Interventions for Walking Activity The literature suggests that the use of walking and physical activity monitoring in daily life is clinically feasible and has significant potential to be employed as a clinical outcome to inform the care and management of children with cerebral palsy. Community-based monitoring with devices reviewed in this chapter can be employed within numerous rehabilitation strategies in this population including but not limited to medications for movement disorders, injection therapy to lower and upper extremity, upper extremity constraint-induced therapy, orthotic management, use of assistive mobility devices, (walker versus crutches) gait training (i.e., overground, treadmill), orthopedic surgery, and neurosurgical interventions (intrathecal baclofen, selective dorsal rhizotomy) as well as lifestyle intervention to enhance physical activity. We will review preliminary and published data describing the application of monitoring devices to describe community-based walking and physical activity in children with CP. All children with cerebral palsy (CP) exhibit some type of movement disorder (i.e., spasticity, dystonia etc.). Often oral medications are employed to decrease the influence of these movement disorders on activities of daily life such as sitting, walking, and upper extremity tasks of daily life. A monitoring device (Actigraph on the wrist) can be employed to document the influence of oral medications on overall excessive movement in a child with upper extremity dystonia during feeding, for example. In this clinical scenario, we would hope to see less extraneous limb movement if the medication is having the desired effect. Similarly, pre-/post-upper extremity injections to optimize upper extremity use for feeding could employ the Actigraph during feeding of one arm. Devices (i.e., three-dimensional accelerometers) on both wrists could be used to document the relative change in bilateral upper extremity use post-constraint-induced therapy treatment. Walking activity in daily life captured by wearable devices can offer valuable information to guide interventions which are employed to optimize walking within the context of the child’s environment. A 2016 clinical pilot study of 11 children with CP and bilateral impairment employed 2 weeks monitoring with the StepWatch to
1026
K.F. Bjornson and N. Lennon
examine the influence of current orthotic prescriptions on community walking (Bjornson et al. 2016). StepWatch data captured strides/day, time walking, as well as walking intensity. Depending on the walking (SW) outcome examined, only two to four of the 11 participants demonstrated improvements with their current orthotic prescription as compared to walking without their orthotics. Recent work has applied StepWatch monitoring pre- and post-short-burst interval locomotor treadmill training (SBLTT) in ambulatory school-aged children with CP (unpublished Bjornson et al. 2016 R21 results). R21 pilot data of pre-/post-SBLTT (n = 12, Figs. 4 and 5) documented enhanced community walking levels (average strides/day +948 strides/day, p < 0.001) at 6 weeks post-SBLTT training. Percent time walking at higher stride/min intensities of medium/high stride (>39 strides/ min) rates increased (+3.8%, p = 0.04) including absolute number of strides/day at medium/high stride rates (+627, p < 0.001). Studies of orthopedic surgery outcomes in lab settings for youth with cerebral palsy (CP) who undergo surgical correction to improve gait reveal positive changes in body structure/function measures such as kinematic and kinetic patterns (Wren et al. 2013) and improvement in activity capacity measures such as gait speed (Gannotti et al. 2007) after surgery. The validation of physical activity (PA) monitors for youth with CP now allows for investigation of performance level outcomes such as change in habitual level of walking activity post-surgery. Lennon and colleagues used the StepWatch to examine the recovery of walking activity in youth who underwent orthopedic surgery to correct gait (Lennon et al. 2015). Preoperative clinical gait analysis at the Nemours duPont Hospital for Children includes collection of habitual walking activity pre- and postoperatively. The surgeons, physiatrists, and physical therapists at the hospital evaluate walking activity data as part of rehabilitation planning during the child’s recovery from
Fig. 4 Average strides/day at baseline, post-SBLTT, and 6 weeks post-SBLTT (n = 12)
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1027
Fig. 5 Percent of total strides/day in low, moderate and high stride rates for 368 typically developing youth (TDY), (Bjornson et al. 2014) 209 youth with CP and 12 youth with CP, Pre/post short-burst LTT. Green- high= >60 stride/min, Red-moderate=30-60 stride/min and Blue- low 0.05) (Lennon et al. 2015). Figure 6 shows recovery trends of both groups relative to expected strides for GMFCS level (Bjornson et al. 2007).
1028
K.F. Bjornson and N. Lennon
Interventions for Physical Activity The Carol and Paul Hatfield Cerebral Palsy Sports and Rehabilitation Center at The St. Louis Children’s Hospital provide children with cerebral palsy and related childhood disabilities the opportunity to experience sport activities in fun, social environments. Known as “Camp Independence,” it is a summer program of intensive physical activities in a sports camp format. Participants are assisted by physical therapists, nurses, aides, and volunteers to play many different sporting activities. Campers spend 7 h a day for up to 4 weeks participating in sport activities such as tennis, swimming, yoga, martial arts, basketball, soccer, cycling, baseball, and dance. Each sport is adapted as necessary to meet the needs of the child. During the summer of 2014, Miros and colleagues recruited 34 campers to wear a FitBit Flex activity tracking device for 2 weeks before camp, while they were attending camp, and a minimum of 2 weeks after camp (St. Louis Childrens Hospital 2015). All activity data were collected using the FitBit app and the fitbit.com website (FitBit 2016). Campers also completed a number of standardized physical outcome measures at the beginning and end of camp to evaluate the functional benefits of camp. Participants were between the ages of 7 and 18 and ranged in physical ability from independently being able to run to requiring a power wheelchair for community mobility. Although there were no significant changes in standardized physical outcome measures, participants were more active on camp days versus non-camp days. The participants averaged 8,693 steps on days they attended camp and 5,730 steps on days they did not attend camp. Activity levels decreased after camp, which families attribute to the lack of community physical activity opportunities accommodating children with CP (St. Louis Childrens Hospital 2015). Active video games (AVGs) are another promising strategy to provide access to fun, active recreation for youth with CP. O’Neil and colleagues examined PA intensity levels in 57 youth with CP (mean = 12 years) at GMFCS levels I (28), II (16), and III (13) during 5 min of play on the X-Box 360 Kinect™ Adventures: (River Rush and Space Pops) games (O’Neil et al. 2916). Data was collected utilizing Polar heart rate monitors, the Cosmed K4b2 indirect calorimeter device, the OMNI rate of perceived exertion (RPE) scale and the Actigraph SW, and BMSW activity monitors. They found that youth played AVGs at light-to-moderate PA intensity. Median MET values were River Rush = 3.0, IQR = 2.5–3.8, and Space Pops = 3.3, IQR = 2.6–3.8. Median OMNI RPE values were River Rush = 3.8, IQR = 1–4, and Space Pops = 4.0 (IQR = 2–6). Findings suggest that commercial AVGs may provide opportunities to promote PA in ambulatory youth with CP but are limited in their ability to promote higher levels of PA intensity and longer bouts of PA (O’Neil et al. 2916, 2016). Researchers and entrepreneurs are now working together to develop custom gaming platforms to meet the needs of youth with physical disabilities. A game that allows physical therapists to change the speed and intensity of the game through a back-end portal allows the therapist to structure game conditions to reach the heart rate and level of physical activity intensity individuals most benefit from. At the same time, the game system can collect data on a player’s progress over a session,
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1029
several sessions, or several weeks. The flexibility of gaming parameters allows the therapist to adjust the game to address therapeutic goals. Having the players wear heart rate monitors and activity monitors during game play allows the therapist to see if the child is reaching the aerobic and activity levels that are health promoting. To date, O’Neil’s pilot study has enrolled 12 children and youth to play three newly developed 20-min games (O’Neil et al. 2016, unpublished work). Youth have enjoyed playing, which is an important factor. Having a fun way to keep kids active has the potential to mitigate. In 2014, van Wely and colleagues conducted a randomized trial examining the effect of 6-month physical activity stimulation program (van Wely et al. 2010, 2014). The investigators employed the StepWatch to capture habitual walking activity. No significant difference was documented in walking level or intensity between intervention groups at 6 and 12 months post. A randomized control trial of an internetbased physical activity intervention was completed by Maher and colleagues in 2010 (Maher et al. 2010). Employing a waist-mounted pedometer (NL1000), no significant differences between intervention and comparison group were documented for number of steps taken per week or self-reported physical activity at 10 and 20 weeks post.
Considerations and Limitations of Walking and Physical Activity Monitoring When selecting a monitor or device to capture WA or PA, there are several factors to consider. First, it is important to select a device that is validated and most directly related to the behavior you are aiming to capture and/or influence with intervention. For example, for walking activity post-lower extremity surgical intervention, you may choose the StepWatch, while physical activity after a home-based AVG program maybe most appropriately measured with the Actigraph. Second, the costs of these devices vary from $30 to $500 per device, and this may or may not include the processing software to access meaningful information to interpret. Third, the ease of processing and getting to outputs that can be readily interpreted should be considered as well. The addition of bluetooth versions (wireless/cloud-based) of devices will now make downloading relatively easier (i.e., StepWatch and Actigraph). Lastly, in a clinical setting, billing for the assessment of walking and/or physical activity can be included within a clinical gait or mobility evaluation and/or a post-intervention outcome assessment physical therapy charge by a physical therapist. The emerging data on WA and PA in children and youth with CP reveals high variability among many of the study samples. Thus, it is important that this variability is taken into account in developing study designs/sample sizes to insure that work is adequately powered to capture clinically meaningful as well as statistical significance. For WA activity in children with CP, a potential “minimal clinical important difference” (MCID) benchmark could be the mean difference between walking activity between GMFCS levels. For example, enroll a sample size adequate
1030
K.F. Bjornson and N. Lennon
to capture the mean difference between WA of children with CP at GMFCS level I versus II (based on current published WA levels) (Bjornson et al. 2014). PA monitors have potential weaknesses that should be considered with their implementation as well. The accuracy and precision of instruments worn on the waist or upper arm can be limited for certain types of upright behaviors that have a low ambulatory component and that may involve upper body work. Similarly WA devices worn on the wrist may not adequately capture the abnormal walking patterns of children with CP and/or after interventions (i.e., orthotics, orthopedic surgery). Recent efforts to enhance the ability of physical activity monitors to capture these behaviors include more densely sampled data and more sophisticated prediction equations. Information about the location or purpose of individual activities is limited, unless information from other sources is integrated with information from the monitor. Technological solutions synchronizing GPS with accelerometers previously employed in populations without physical disability are being examined in children with CP (Bjornson et al. 2016, unpublished data).
Conclusions/Summary In summary, accelerometers have been validated and reliably employed to document walking and physical activity of children with cerebral palsy across functional levels. Relative to walking activity, published information to date offers reference information for walking activity levels, walking patterns, and intensity of walking by GMFCS levels as compared to children without motor limitations with the StepWatch. Physical activity has been documented for ambulatory and non-ambulatory children with CP with several devices (i.e., Actigraph, BodyMedia, etc.) and validated to energy cost in ambulatory children. Cut points for interpretation of PA intensity with the Actigraph in children with CP of varying ages are emerging in the literature. Based on the body of knowledge reviewed in this chapter, accelerometry can be effectively employed to capture walking and physical activity in children and youth with CP. This knowledge should inform clinical and/or research questions relative to performance by functional levels, documenting natural history and directing rehabilitation strategies as ecologically based outcomes of WA and PA within the context of daily life. Future development of WA and PA monitoring in children with CP should focus on the capturing upper extremity functional activities, expand the use of bluetooth technology and/or cloud-based capture of data and synchronization accelerometry with GPS and patient/parent reported qualitative outcomes.
Cross-References ▶ Assessing Clubfoot and Cerebral Palsy by Pedobarography ▶ Assessing the Impact of Aerobic Fitness on Gait ▶ Clinical Gait Assessment by Video Observation and 2D Techniques
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1031
▶ Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy ▶ EMG Activity in Gait: The Influence of Motor Disorders ▶ Foot and Ankle Motion in Cerebral Palsy ▶ Functional Effects of Foot Orthoses ▶ Gait Scores: Interpretations and Limitations ▶ Interpreting Spatiotemporal Parameters, Symmetry, and Variability in Clinical Gait Analysis ▶ Natural History of Cerebral Palsy and Outcome Assessment ▶ Oxygen Consumption in Cerebral Palsy ▶ Spasticity Effect in Cerebral Palsy Gait ▶ Strength Related Stance Phase Problems in Cerebral Palsy ▶ Swing Phase Problems in Cerebral Palsy
References Alotaibi M, Long T, Kennedy E, Bavishi S (2014) The efficacy of GMFM-88 and GMFM-66 to detect changes in gross motor function in children with cerebral palsy (CP): a literature review. Disabil Rehabil 36:617–627 Andreacci JL, Dixon C, Dube JJ, McConnell TR (2007) Validation of SenseWear Pro2 armband to assess energy expenditure during treadmill exercise in children 7–10 years of age. J Exerc Physiol Online 10:35–42 Bagley AM, Gorton GE, Bjornson K, Bevans K, Stout JL, Narayanan U, Tucker CA (2011) Factorand item-level analyses of the 38-item activities scale for kids–performance. Dev Med Child Neurol 53:161–166 Barreira TV, Katzmarzyk PT, Johnson WD, Tudor-Locke C (2012) Cadence patterns and peak cadence in US children and adolescents: NHANES, 2005–2006. Med Sci Sports Exerc 44:1721–1727 Bassett DR, Dinesh J (2010) Use of pedometers and accelerometers in clinical populations: validity and reliability issues. Phys Ther Rev 15(3):135–142 Beckung E, Hagberg G (2002) Neuroimpairments, activity limitations, and participation restrictions in children with cerebral palsy. Dev Med Child Neurol 44:309–316 Beckung E, Carlsson G, Carlsdotter S, Uvebrant P (2007) The natural history of gross motor development in children with cerebral palsy aged 1 to 15 years. Dev Med Child Neurol 49:751–756 van den Berg-Emons HJ, Saris WH, de Barbanson DC, Westerterp KR, Huson A, van Baak MA (1995) Daily physical activity of school children with spastic diplegia and of healthy control subjects. J Pediatr 127:578–584 Bjornson KF, Belza B, Kartin D, Logsdon R, McLaughlin JF (2007) Ambulatory physical activity performance in youth with cerebral palsy and youth who are developing typically. Phys Ther 87:248–257 Bjornson K, Song K, Lisle J, Robinson S, Killien E, Barrett T, Zhou C (2010) Measurement of walking activity throughout childhood: influence of leg length. Pediatr Exerc Sci 22:581–595 Bjornson KF, Song K, Zhou C, Coleman K, Myaing M, Robinson SL (2011) Walking stride rate patterns in children and youth. Pediatr Phys Ther 23:354–363 Bjornson KF, Zhou C, Stevenson RD, Christakis D (2014a) Relation of stride activity and participation in mobility-based life habits among children with cerebral palsy. Arch Phys Med Rehabil 95:360–368 Bjornson K, Zhou C, Stevenson RD, Christakis D, Song K (2014b) Walking activity patterns in youth with cerebral palsy and youth developing typically. Disabil Rehabil 36:1279–1284
1032
K.F. Bjornson and N. Lennon
Bjornson K, Zhou C, Fatone S, Orendurff M, Stevenson R, Rashid S (2016) The effect of ankle-foot orthoses on community-based walking in cerebral palsy: a clinical pilot study. Pediatr Phys Ther 28:179–186 Bohannon R (1997) Comfortable and maximum walking speed of adults aged 20–79 years: reference values and determinants. Age Ageing 26:15–19 Bürgi R, Tomatis L, Murer K, de Bruin ED (2015) Localization of physical activity in primary school children using accelerometry and global positioning system. PLoS One 10:e0142223 Bussmann JBJ, Ebner-Priemer U, Fahrenberg J (2009) Ambulatory activity monitoring: progress in measurement of activity, posture, and specific motion patterns in daily life. Eur Psychol 14:142–152 Cain K, Salis JF, Conway TL, van Dyck D, Calhoon L (2013) Using accelerometers in youth physical activity studies: a review of methods. J Phys Act Health 10:437–450 Capio CM, Sit CHP, Abernethy B, Masters RSW (2012) Fundamental movement skills and physical activity among children with and without cerebral palsy. Res Dev Disabil 33:1235–1241 Carlson JA, Schipperijn J, Kerr J, Saelens BE, Natarajan L, Frank LD, Glanz K, Conway TL, Chapman JE, Cain KL, Sallis JF (2016) Locations of physical activity as assessed by GPS in young adolescents. Pediatrics 137:e20152430 Caspersen CJ, Powell KE, Christenson GM (1985) Physical activity, exercise, and physical fitness: definitions and distinctions for health-related research. Public Health Report 100:126–131 Cesari M, Kritchevsky SB, Penninx BWHJ, Nicklas BJ, Simonsick EM, Newman AB, Tylavsky FA, Brach JS, Satterfield S, Bauer DC, Visser M, Rubin SM, Harris TB, Pahor M (2005) Prognostic value of usual gait speed in well-functioning older people – results from the health, aging and body composition study. J Am Geriatr Soc 53:1675–1680 Chillón P, Panter J, Corder K, Jones AP, van Sluijs EMF (2015) A longitudinal study of the distance that young people walk to school. Health Place 31:133–137 Chrysagis N, Skordilis EK, Koutsouki D (2014) Validity and clinical utility of functional assessments in children with cerebral palsy. Arch Phys Med Rehabil 95:369–374 Clanchy K, Tweedy S, Boyd R, Trost S (2011) Validity of accelerometry in ambulatory children and adolescents with cerebral palsy. European J Appl Physiol 111(12):2951–9. https://doi.org/ 10.1007/s00421-011-1915-2 Claridge EA, McPhee PG, Timmons BW, Martin Ginis KA, Macdonald MJ, Gorter JW (2015) Quantification of physical activity and sedentary time in adults with cerebral palsy. Med Sci Sports Exerc 47:1719–1726 Corder K, Brage S, Ramachandran A, Snehalatha C, Wareham N, Ekelund U (2007) Comparison of two Actigraph models for assessing free-living physical activity in Indian adolescents. J Sports Sci 25:1607–1611 Del Pilar Duque Orozco M, Abousamra O, Church C, Lennon N, Henley J, Rogers KJ, Sees JP, Connor J, Miller F (2016) Reliability and validity of Edinburgh visual gait score as an evaluation tool for children with cerebral palsy. Gait Posture 49:14–18 Demant Klinker C, Schipperijn J, Toftager M, Kerr J, Troelsen J (2015) When cities move children: development of a new methodology to assess context-specific physical activity behaviour among children and adolescents using accelerometers and GPS. Health Place 31:90–99 Diggory P, Gorman M, Schwarz J, Helme R (1994) An automatic device to measure time spent upright. Clin Rehabil 8:353–357 Evenson KR, Catellier DJ, Gill K, Ondrak KS, McMurray RG (2008) Calibration of two objective measures of physical activity for children. J Sports Sci 26:1557–1565 FITBIT (2016) Fitbit Products [Online]. Available: https://www.fitbit.com/ Foster RC, Lanningham-Foster LM, Manohar C, McCrady SK, Nysse LJ, Kaufman KR, Padgett DJ, Levine JA (2005) Precision and accuracy of ankle-worn accelerometer-based pedometer in step counting and energy expenditure. Prev Med 41:778–783 Fritz S, Lusardi M (2009) White paper: “walking speed: the sixth vital sign”. J Geriatr Phys Ther 32:46–49
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1033
Gannotti ME, Gorton GEI, Nahorniak MT, Masso PD, Landry B, Lyman J, Sawicki R, Hagedorn K, Ross E, Warner J (2007) Postoperative gait velocity and mean knee flexion in stance of ambulatory children with spastic diplegia four years or more after multilevel surgery. J Pediatr Orthop 27:451–456 Gor-García-Fogeda MD, Cano de la Cuerda R, Carratalá Tejada M, Alguacil-Diego IM, MolinaRueda F (2016) Observational gait assessments in people with neurological disorders: a systematic review. Arch Phys Med Rehabil 97:131–140 Gorter JW, Noorduyn SG, Obeid J, Timmons BW (2012) Accelerometry: a feasible method to quantify physical activity in ambulatory and nonambulatory adolescents with cerebral palsy. Int J Pediatr 2012:6 Gorton GE III, Stout JL, Bagley AM, Bevans K, Novacheck TF, Tucker CA (2011) Gillette functional assessment questionnaire 22-item skill set: factor and Rasch analyses. Dev Med Child Neurol 53:250–255 Graham HK, Harvey A, Rodda J, Nattrass GR, Pirpiris M (2004) The functional mobility scale (FMS). J Pediatr Orthop 24:514–520 Granger CV, Hamilton BB (1993) The uniform data system for medical rehabilitation report of first admissions for 1991. Am J Phys Med Rehabil 72:33–38 Haak P, Lenski M, Hidecker MJC, Li M, Paneth N (2009) Cerebral palsy and aging. Dev Med Child Neurol 51:16–23 Health M (2015) StepWatch [Online]. Available: https://modushealth.com/. Accessed 17 Nov 2015 Healthcare O (2016) Omron Pedometers [Online]. Available: https://omronhealthcare.com/fitness/ activity-trackers-and-pedometers/ Imms C, Reilly S, Carlin J, Dodd K (2008) Diversity of participation in children with cerebral palsy. Dev Med Child Neurol 50:363–369 Iannotti RJ, Wang J (2013) Patterns of physical activity, sedentary behavior, and diet in U.S. adolescents. J Adolescent Health 53(2):280–6. https://www.ncbi.nlm.nih.gov/pubmed/ 23642973 Ishikawa S, Kang M, Bjornson KF, Song K (2013) Reliably measuring ambulatory activity levels of children and adolescents with cerebral palsy. Arch Phys Med Rehabil 94:132–137 Kamp FA, Lennon N, Holmes L, Dallmeijer AJ, Henley J, Miller F (2014) Energy cost of walking in children with spastic cerebral palsy: relationship with age, body composition and mobility capacity. Gait Posture 40:209–214 Kang M, Bjornson KF, Barreira TV, Ragan BG, Song K (2014) The minimum number of days required to establish reliable physical activity estimates in children aged 2–15 years. Physiol Meas 35:2229 Karabulut M, Crouter S, Bassett DR (2005) Comparison of two waist-mounted and two anklemounted electronic pedometers. Eur J Appl Physiol 96:334–335 Keawutan P, Bell K, Davies PSW, Boyd RN (2014) Systematic review of the relationship between habitual physical activity and motor capacity in children with cerebral palsy. Res Dev Disabil 35:1301–1309 Keawutan P, Bell KL, Oftedal S, Davies PSW, Boyd RN (2016) Validation of accelerometer cut-points in children with cerebral palsy aged 4 to 5 years. Pediatr Phys Ther 28:427–434 Kerr C, Parkes J, Stevenson M, Cosgrove AP, McDowell BC (2008) Energy efficiency in gait, activity, participation and health status in children wtih cerebral palsy. Dev Med Child Neurol 50:204–210 Ko J, Kim M (2013) Reliability and responsiveness of the gross motor function measure-88 in children with cerebral palsy. Phys Ther 93:393–400 Koehler K, Abel T, Wallmann-Sperlich B, Dreuscher A, Anneken V (2015) Energy expenditure in adolescents with cerebral palsy: comparison of the SenseWear armband and indirect calorimetry. J Phys Act Health 12:540–545 Kozey SL, Staudenmayer JW, Troiano RP, Freedson PS (2010) Comparison of the ActiGraph 7164 and the ActiGraph GT1M during self-paced locomotion. Med Sci Sports Exerc 42:971–976
1034
K.F. Bjornson and N. Lennon
Lee J-M, Kim Y, Welk GJ (2014) Validity of consumer-based physical activity monitors. Med Sci Sports Exerc 46:1840–1848 Lee J-M, Kim Y, Bai Y, Gaesser GA, Welk GJ (2016) Validation of the SenseWear mini armband in children during semi-structure activity settings. J Sci Med Sport 19:41–45 Lennon N, Hulbert R, Church C, Miller F (2015) Surgical burden and recovery of walking performance in youth with cerebral palsy. Dev Med Child Neurol 57:96–97 Maher CA, Williams MT, Olds TIM, Lane AE (2010) An internet-based physical activity intervention for adolescents with cerebral palsy: a randomized controlled trial. Dev Med Child Neurol 52:448–455 McCrorie PRW, Fenton C, Ellaway A (2014) Combining GPS, GIS, and accelerometry to explore the physical activity and environment relationship in children and young people – a review. Int J Behav Nutr Phys Act 11:93 McDonald CM, Widman L, Abresch T, Walsh S, Walsh D (2005) Utility of a step activity monitor for the measurement of daily ambulatory activity in children. Arch Phys Med Rehabil 86:793–801 Melanson EL, Knoll JR, Bell ML, Donahoo WT, Hill JO, Nysse LJ, Lanningham-Foster L, Peters JC, Levine JA (2004) Commercially available pedometers: considerations for accurate step counting. Prev Med 39:361–368 Mitchell LE, Ziviani J, Boyd RN (2015) Habitual physical activity of independently ambulant children and adolescents with cerebral palsy: are they doing enough? Phys Ther 95:202 Mitre N, Lanningham-Foster L, Foster R, Levine JA (2009) Pedometer accuracy in children: can we recommend them for our obese population? Pediatrics 123:e127–e131 Morris C, Kurinczuk JJ, Fitzpatrick R, Rosenbaum PL (2006) Do the abilities of children with cerebral palsy explain their activities and participation? Dev Med Child Neurol 48:954–961. Dev Med Child Neurol. 2007 Feb;49(2):122 Nicholson K, Lennon N, Hulbert R, Church C, Miller F (2017) Pre-operative walking activity in youth with cerebral palsy. Res Dev Disabil 60:77–82 Nooijen CF, de Groot JF, Stam HJ, van den Berg-Emons RJ, Bussmann HB (2015) Validation of an activity monitor for children who are partly or completely wheelchair-dependent. J Neuroeng Rehabil 12:11 O’Neil ME, Fragala-Pinkham MA, Forman JL, Trost SG (2014) Measuring reliability and validity of the ActiGraph GT3X accelerometer for children with cerebral palsy: a feasibility study. J Pediatr Rehabil Med 7:233–240 O’Neil ME, Fragala-Pinkham M, Lennon N, George A, Forman J, Trost SG (2016) Reliability and validity of objective measures of physical activity in youth with cerebral palsy who are ambulatory. Phys Ther 96:37–45 O’Neil ME, Fragala-Pinkham M, Lennon N, Trost SG (2916) Active video games to promote physical activity in youth with cerebral palsy. In: IV STEP 2016. Columbus: Pediatric and Neurology Sections, American Physical Therapy Association Oeffinger D, Bagley A, Rogers S, Gorton G, Kryscio R, Abel M, Damiano D, Barnes D, Tylkowski C (2008) Outcome tools used for ambulatory children with cerebral palsy: responsiveness and minimum clinically important differences. Dev Med Child Neurol 50:918–925 Oftedal S, Bell KL, Davies PSW, Ware RS, Boyd RN (2014) Validation of accelerometer cut points in toddlers with and without cerebral palsy. Med Sci Sports Exerc 46:1808–1815 Oliver M, Badland H, Mavoa S, Duncan MJ, Duncan S (2010) Combining GPS, GIS, and accelerometry: methodological issues in the assessment of location and intensity of travel behaviors. J Phys Act Health 7:102–108 Palisano R, Rosenbaum P, Bartlett D, Livingston MH (2007) GMFCS- R & E gross motor function classification system expanded and revised, 2007. CanChild Centre for Childhood Disability, McMasters University, Hamilton Pirpiris M, Graham HK (2004) Uptime in children with cerebral palsy. J Pediatr Orthop 24:521–528 Play G (2016) Virtual walk treadmill or GPS [Online]. Available: https://play.google.com/store/ apps/details?id=com.virtualwalk.android&hl=en
Walking and Physical Activity Monitoring in Children with Cerebral Palsy
1035
Portney LW, Watkins MP (2009) Foundations of clinical research: applications to practice. Pearson/ Prentice Hall, Upper Saddle River Postma K, van den Berg-Emons HJG, Bussmann JBJ, Sluis TAR, Bergen MP, Stam HJ (2005) Validity of the detection of wheelchair propulsion as measured with an activity monitor in patients with spinal cord injury. Spinal Cord 43:550–557 Puyau MR, Adolph AL, Vohra FA, Butte NF (2002) Validation and calibration of physical activity monitors in children. Obes Res 10:150–157 Puyau MR, Adolph AL, Vohra FA, Zakeri I, Butte NF (2004) Prediction of activity energy expenditure using accelerometers in children. Med Sci Sports Exerc 36:1625–1631 Rothney MP, Apker GA, Song Y, Chen KY (2008) Comparing the performance of three generations of ActiGraph accelerometers. J Appl Physiol 105:1091–1097 Ryan JM, Walsh M, Gormley J (2014) A comparison of three accelerometry-based devices for estimating energy expenditure in adults and children with cerebral palsy. J Neuroeng Rehabil 11:116 Services, U. S. D. O. H. A. H. (2012) Physical activity guidelines for Americans: fact sheet for professionals [Online]. Available: http://www.health.gov/paguidelines/factsheetprof.aspx. Accessed 06 Feb 2013 Sirard JR, Trost SG, Pfeiffer KA, Dowda M, Pate RR (2005) Calibration and evaluation of an objective measure of physical activity in preschool children. J Phys Act Health 3:345–357 Slaman J, Dallmeijer A, Stam H, Russchen H, Roebroeck M, van den Berg-Emons R (2013) The six-minute walk test cannot predict peak cardiopulmonary fitness in ambulatory adolescents and young adults with cerebral palsy. Arch Phys Med Rehabil 94:2227–2233 Song KM, Bjornson KF, Capello T, Coleman K (2006) Use of the StepWatch activity monitor for characterization of normal activity levels in children. J Pediatr Orthop 26:245–249 St Louis Childrens Hospital (2015) Fitbit fun: does Camp Independence increase activity of chil dren with cerebral palsy? Pediatr Perspect [Online], Fall Available: http://www.stlouischildrens. org/sites/default/files/health_professionals/images/PedPerspective_Fall2015web.pdf. Accessed 11 Dec 2016 Sullivan E, Barnes D, Linton JL, Calmes J, Damiano DL, Oeffinger D, Abel M, Bagley A, Gorton G, Nicholson D, Rogers S, Tylkowski C (2007) Relationship among functional outcome measures used for assessing children with ambulatory CP. Dev Med Child Neurol 49:338–344 Thomas SS, Buckon CE, Piatt JH, Aiona MD, Sussman MD (2004) A 2-year follow-up of outcomes following orthopedic surgery or selective dorsal rhizotomy in children with spastic diplegia. J Pediatr Orthop B 13:358–366 Thompson P, Beath T, Bell J, Jacobson G, Phair RSNM, Wright FV (2008) Test-retest reliability of the 10-metre fast walk test and 6 minute walk test in ambulatory school-aged children with cerebral palsy. Dev Med Child Neurol 50:370–376 Trost SG (2007) State of the art reviews: measurement of physical activity in children and adolescents. Am J Lifestyle Med 1:299–314 Trost SG, McIver KL, Pate RR (2005) Conducting accelerometer-based activity assessments in field-based research. Med Sci Sports Exerc 37:S531–S543 Trost SG, Way R, Okely AD (2006) Predictive validity of three Actigraph energy expenditure equations for children. Med Sci Sports Exerc 38:380–387 Trost SG, Fragala-Pinkham M, Lennon N, O’Neil ME (2016) Decision trees for detection of activity intensity in youth with cerebral palsy. Med Sci Sports Exerc 48:958–966 Tudor-Locke CE, Myers AM (2001) Methodological considerations for researchers and practitioners using pedometers to measure physical (ambulatory) activity. Res Q Exerc Sports 72:1–12 Tudor-Locke C, Williams JE, Reis JP, Pluto D (2002) Utility of pedometers for assessing physical activity: convergent validity. [Review] [49 refs] Sports Med 32:795–808 Tudor-Locke C, Craig C, Beets M, Belton S, Cardon G, Duncan S, Hatano Y, Lubans D, Olds T, Raustorp A, Rowe D, Spence J, Tanaka S, Blair S (2011a) How many steps/day are enough? For children and adolescents. Int J Behav Nutr Phys Act 8:78
1036
K.F. Bjornson and N. Lennon
Tudor-Locke C, Craig CL, Brown WJ, Clemes SA, de Cocker K, Giles-Corti B, Hatano Y, Inoue S, Matsudo SM, Mutrie N, Oppert J-M, Rowe DA, Schmidt MD, Schofield GM, Spence JC, Teixeira PJ, Tully MA, Blair SN (2011b) How many steps/day are enough? For adults. Int J Behav Nutr Phys Act 8:79–79 Tudor-Locke C, Brashear M, Katzmarzyk PT, Johnson WD (2012) Peak stepping cadence in freeliving adults: 2005–2006 NHANES. J Phys Act Health 9:1125–1129 Tyron WW (1991) Activity measurement in psychology and medicine. Plenum Press, New York U.S. Dept of Health and Human Services (2008) Physical activity guidelines for Americans summary [Online]. Available: http://www.health.gov/PAGuidelines/. Accessed 30 Nov 2016 Voorman JM, Dallmeijer AJ, Schuengel C, Knol DL, Lankhorst G, Becher JG (2006) Activities and participation of 9- to 13-year-old children with cerebral palsy. Clin Rehabil 20:937–948 Warms CA, Whitney JD, Belza B (2008) Measurement and description of physical activity in adult manual wheelchair users†. Disabil Health J 1:236–244 van Wely L, Becher J, Reinders-Messelink H, Lindeman E, Verschuren O, Verheijden J, Dallmeijer A (2010) LEARN 2 MOVE 7-12 years: a randomized controlled trial on the effects of a physical activity stimulation program in children with cerebral palsy. BMC Pediatr 10:77 van Wely L, Becher JG, Balemans ACJ, Dallmeijer AJ (2012) Ambulatory activity of children with cerebral palsy: which characteristics are important? Dev Med Child Neurol 54:436–442 van Wely L, Balemans ACJ, Becher JG, Dallmeijer AJ (2014) Physical activity stimulation program for children with cerebral palsy did not improve physical activity: a randomised trial. J Physiother 60:40–49 Wen LM, van der Ploeg HP, Kite J, Cashmore A, Rissel C (2010) A validation study of assessing physical activity and sedentary behavior in children aged 3 to 5 years. Pediatr Exerc Sci 2:408–420 Wieters KM, Kim J, Lee C (2012) Assessment of wearable global positioning system units for physical activity research. J Phys Act Health 9:913–923 Wilson NC, Signal N, Naude Y, Taylor D, Stott NS (2015) Gait deviation index correlates with daily step activity in children with cerebral palsy. Arch Phys Med Rehabil 96:1924–1927 Wilson NC, Mudge S, Stott NS (2016) Variability of total step activity in children with cerebral palsy: influence of definition of a day on participant retention within the study. BMC Res Notes 9:411 Wininger M, Bjornson K (2016) Filtering for productive activity changes outcomes in step-based monitoring among children. Physiol Meas 37(12):2231 Wood E, Rosenbaum P (2000) The gross motor function classification system for cerebral palsy: a study of reliability and stability over time. Dev Med Child Neurol 42:292–296 World Health Organization (2002) International classification of functioning, disability and health (ICF). World Health Organization, Geneva Wren TAL, Lening C, Rethlefsen SA, Kay RM (2013) Impact of gait analysis on correction of excessive hip internal rotation in ambulatory children with cerebral palsy: a randomized controlled trial. Dev Med Child Neurol 55:919–925 YAMAX-DIGIWALKER.COM (2016) Yamax-Digiwalker [Online]. Warminster. Available: https://www.yamax-digiwalker.com/
Spasticity Effect in Cerebral Palsy Gait Marlene Cristina Neves Rosa and André Gonçalo Gomes Roque
Abstract
A high number of children with cerebral palsy (CP) have spastic gait and consequently abnormalities in joint patterns. Several factors have been contributing to the lack of consensus on the spasticity effect in cerebral palsy gait and would be summarized and discussed in this chapter, e.g., spastic gait patterns are in constant evolution during the process of growth; there are still considerable limitations in the methodologies used to assess spasticity during gait; a wide range of rehabilitation strategies have been explored to control spasticity during gait. Spastic gait patterns are divided in hemiplegic (5 types) and diplegic (4 types) with the most prevalent joint abnormalities described in the sagittal plane. Ashworth, Tardieu, and DAROM scales and Pendulum Tests are widely used to assess spasticity but do not reliably explain the spasticity effects during gait. Orthotics, adequate exercise or handling techniques, botulinum injections, or surgical procedures have been used to manage spasticity effects. Keywords
Cerebral palsy • Gait • Spasticity
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1038 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1039 Spastic Gait Disorders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1039
M.C.N. Rosa (*) Piaget Institute, Viseu, Portugal e-mail: [email protected] A.G.G. Roque (*) Physiotherapy, University of Averio, Aveiro, Portugal e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_55
1037
1038
M.C.N. Rosa and A.G.G. Roque
Methodologies to Assess Spasticity During Gait in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . Spasticity Treatments with Impact in CP Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1043 1046 1050 1050
Introduction Spasticity is one of the sequelae of neuromuscular disabilities, affecting 260–340 per 100,000 children with cerebral palsy (Dana and Cub 2013). Spasticity is the result of pyramidal tract lesions, but there are several theories to explain the neurophysiologic mechanisms related with this phenomenon (Dana and Cub 2013). The most commonly used definition of spasticity is that of Lance (1980) that explains it as “a velocity-dependent increase in tonic stretch reflexes (muscle tone) with exaggerated tendon jerks. . .” (Hobart et al. 1994). It depends on overexcitability of spinal alpha motor neurons, as a consequence of the interruption of descending modulatory carried by the corticospinal, vestibulospinal, and reticulospinal tracts (Filloux 1996). In specific, spasticity can be explained by the following pathophysiologic mechanisms: reduced reciprocal inhibition of antagonist motor neuron pools by Ia afferents, decreased presynaptic inhibition of Ia afferents, and decreased nonreciprocal inhibition by Ib terminals (Hobart et al. 1994). Despite this lack of consensus explaining the neurophysiologic mechanisms underlying spasticity, it is one of the most common disabling consequences in cerebral palsy, causing contractures, joint subluxations, and, indirectly, fatigue, loss of dexterity and coordination, and balance disorders (Dana and Cub 2013). As cerebral palsy (CP) is a life-lasting condition, spasticity also affects the growth of the individual, hampering both muscle and skeletal development and resulting in symmetric or asymmetric biomechanical deficiencies depending on the PC type (Bar-On et al. 2015). Also, one has to consider that CP originates in the infancy or childhood, and the consequences of spasticity have to be separated from that resulting from adult lesions. In the second case, motor control has been normally developed, as opposed by the first case in which spasticity-related gait alterations will direct the motor learning lifelong. These consequences have serious impact in patients´ functionality, specifically in gait. Gait disorders due to spasticity can assume different patterns, depending on the muscles affected, which may cause different effects in joint kinematics (Miller 2004). The first section (“Spastic Gait Disorders”) of this chapter will summarize the types of spastic gait disorders and its characteristics, according to the muscles affected. In addition, efficient methodologies to assess spasticity are crucial to understand its effect in gait pattern (Scholtes et al. 2006). Most of the instruments described in the literature are not able to assess spasticity under rhythmic motor tasks such as gait (Scholtes et al. 2006). Alternative methodologies (e.g., electromyography) have critical aspects that would be discussed in this chapter as they are not aligned with the original definition of spasticity. Section two (“Methodologies to Assess Spasticity During Gait in Cerebral Palsy”) will summarize weaknesses and strengths of
Spasticity Effect in Cerebral Palsy Gait
1039
different scales and instruments that have been used to assess spasticity during gait or the related phenomenon. Section three (“Spasticity Treatments with Impact in CP Gait”) of this chapter will give a simple perspective of the widely accepted approaches to manage spasticity and its effects on CP gait, not focusing on specific results or merits of each.
State of the Art The knowledge of CP spastic gait patterns is well established and despite some children may present with mixed patterns, often they walk in a typical bilateral equinus. Also spasticity contribution to the different gait patterns is defined in the literature; however, the specific association between spasticity and different components and phases of gait is only recently subjected of throughout investigation and will not therefore be exploited in this chapter. This fact is related to the newly available research methods that allows to study spasticity during gait but also to the capacity to establish associations between clinical and research measures of CP gait and spasticity. The methods and therapeutics that address spasticity and CP gait are numerous; some of them are well established for a long time with different degrees of success (orthopedic surgery, stretching, botulinum toxin A, orthotics), while some are more recent yet promising (selective dorsal rhizotomy, intrathecal baclofen, virtual reality, and transcranial magnetic stimulation). The choice between different approaches is a truly clinical decision, which should be based on CP gait analysis, the available evidence, and the individual presented to evaluation. The only conclusion that our readers should obtain toward the treatment theme is that a throughout evaluation of the characteristics of CP gait is fundamental to the success of each of the selected interventions. As such, it is not our objective in this chapter to demonstrate the results of the different approaches but to indicate the existing managing possibilities and, when they were studied, the expected gait outcomes of those interventions and possible implications for the future of the adolescent and adult with CP.
Spastic Gait Disorders Cerebral palsy (CP) is traditionally classified by the clinical type (motor dysfunctional) and topography (anatomical region of the lesion). The motor dysfunctional patterns are classified as (i) spastic, which is the most common pattern, and (ii) ataxic, hypotonic, dyskinetic, and mixed (Morais Filho et al. 2014). Approximately two thirds of all patients with CP suffer from spasticity (Awaad and Rizk 2012). The most affected muscles are the antigravity muscles. Commonly, and as a result of this spasticity, legs are in extended and adducted position and arms are in flexed, internal rotation and pronated position (Awaad and Rizk 2012). In gait, several deviations have been reported as a consequence of spasticity in hemiplegic and diplegic patients, most of them based on joint deviations in the sagittal plane (O’Byrne et al. 1998). Ultimately, these deviations reduce walking
1040
M.C.N. Rosa and A.G.G. Roque
speed and increase energy expenditure, which cause functional inability (Piccinini et al. 2007; Novacheck et al. 2000). Spastic motor deviations are consistent from stride to stride and day to day (Rodda and Graham 2001). Therefore, detectable changes are generally seen as a result of an intervention or as a change related with the aging process and therefore should be monitored during the rehabilitation process (Rodda and Graham 2001). There are at least four different gait patterns in hemiplegic patients with CP all characterized by more distal involvement, which means that true equinus is the basis of the most common gait hemiplegic patterns. In hemiplegia type 1, the most prominent characteristic is drop foot during swing phase caused by inability to control dorsiflexors. Two subtypes characterize hemiplegia type 2. The subtype 2.A includes equinus, neutral knee, and extended hip, while subtype 2.B is characterized by equinus, recurvatum knee, and extended hip. This is the most common type found in clinical practice and it is observed in the stance phase (spasticity in gastrocnemius). Drop foot is also noted during the swing phase because tibialis anterior is impaired. Associated with the predominance of ankle plantarflexion, the knee may adopt the position of recurvatum or extension. The main characteristics of the hemiplegic gait type 3 is the impaired ankle flexion in swing due to gastrocnemius spasticity or contracture and the “stiff knee gait” because of co-contraction between hamstrings and quadriceps. Finally, the hemiplegia type 4 is similar to the pattern of spastic diplegia (but unilateral, therefore asymmetric) with increasing proximal involvement. The characteristics in the sagittal plane are flexed knee, flexed hip, and anterior pelvic tilt. Two other planes present abnormalities in lower limb posture. In the coronal plane, there is hip adduction and in the transverse plane, there is internal rotation. Table 1 summarizes gait hemiplegic patterns in people with cerebral palsy (Rodda and Graham 2001). In contrast with hemiplegic gait, diplegic gait is characterized by more proximal involvement, which means that apparent equinus and crouch gait are very common (Rodda et al. 2004). The five different diplegic gait patterns are schematically described in Table 2. In the up-down direction across the table, there is a decreasing equinus, an increasing proximal involvement, and a change in the direction of the ground reaction force, i.e., from the front of the knee to behind it. The key muscles in these patterns are hip and knee flexors and ankle plantarflexors. These deformities in sagittal gait plane are probably responsible for others found in transverse plane such as torsional deformities of the femur and tibia. The true equinus hemiplegic gait pattern (type 1) is one of the most incident patterns and it is characterized by an equinus and a fully extended knee in late stance. About the jump gait (type II), it is characterized by hip and knee flexion during all stance phase, and then, in late stance, the main characteristic is the equinus. These deformities give the appearance that the individual is jumping up and down, which explains the identification of this gait pattern. In apparent equinus (type III), there are knee and hip flexion contractures that explain why the heel is not in contact with the ground, despite it is not a true equinus foot. In the crouch gait (type IV), three simultaneous conditions are confirmed: hip and knee flexed and ankle kinematics in calcaneus range. Finally, each lower limb presents a different kinematics pattern in the asymmetrical gait
Spasticity Effect in Cerebral Palsy Gait
1041
Table 1 Characteristics of gait patterns in spastic cerebral palsy hemiplegia (Rodda and Graham 2001) Spastic hemiplegia Type I Drop foot
Joint deviation Drop foot
Type II 2.a True equinus 2.b. True equinus/ recurvatum knee
Equinus + neutral knee + extended hip Equinus + recurvatum knee + extended hip Ankle dorsiflexion impaired Flexed stiff knee
Type III True equines/ jump knee
Type IV Equinus/ jump knee Pelvis rotation/hip flexion/ adducted/ internal rotation
Equinus + flexed stiff knee + flexed hip + anterior pelvic tilt Hip adduction Hip internal rotation
Characteristics Plane of Problematic movement gait phase Sagittal SW
Origin of the problem Loss of selective control in dorsiflexors No calf contractures Spasticity/ contracture of gastrocnemius/ soleus muscles
Sagittal
ST
Sagittal
SW
Spasticity/ contracture of gastrocnemius/ soleus muscles Hamstring/ quadricep co-contraction
Sagittal Coronal Transverse
ST SW
Spasticity/ contracture of gastrocnemius/ soleus muscles Hamstring/ quadricep co-contraction Spasticity/ contracture of hip adductors and hip flexors
Management Ankle foot orthosis
Botulinum toxin type A (spasticity) Orthotic support (mild contracture) Tendo Achilles + calf lengthening (fixed contractures) Botulinum toxin type A (spasticity) Tendo Achilles + hamstring lengthening (contractures) Orthotic support (plantar flexion + knee extension couple) Botulinum toxin type A (spasticity) Tendo Achilles + hamstrings + hip adductors + iliopsoas lengthening (contractures) Orthotic support (plantar flexion + knee extension couple)
(type V), i.e., one lower limb is usually classified as belonging to group III and the other is considered as belonging to group II (Rodda et al. 2004). Joint problems may be primary or secondary, with the related problem in the joint or elsewhere. Secondary joint adjustments can be spontaneously reversed with the solving of the primary affection. However, if left untreated, it can become a primary problem that progresses into adulthood. Usually asymmetrical gait has one lower limb toe walking due to spasticity; however, if the child is strong enough, the tendency is for gait symmetry. Thus, the contralateral lower limb will also toe
1042
M.C.N. Rosa and A.G.G. Roque
Table 2 Characteristics of gait patterns in spastic cerebral palsy diplegia (Rodda et al. 2004) Spastic diplegia Type I True equinus Type II Jump gait
Type III Apparent equinus Type IV Crouch gait
Type V Asymmetric gait
Joint deviation Anterior/normal pelvic tilt + recurvatum/normal knee position + drop foot Anterior/normal pelvic tilt + knee and hip from excessively flexed to extended Anterior/normal pelvic tilt + knee and hip flexed Posterior/normal pelvic tilt + knee and hip flexed + foot in excessive dorsiflexion One lower limb – pattern III The other – pattern II
Characteristics Plane of Problematic movement gait phase Sagittal Stance phase
Origin of the problem Contracture of the hamstrings Contracture of calf muscles Flexion contractures of knee and hip
Sagittal
Stance phase
Sagittal
Stance phase
Hamstrings and psoas contraction
Sagittal
Stance phase
Hamstrings and psoas contraction
Sagittal
Stance phase
Lower limb (pattern II) – Flexion contractures of knee and hip Lower limb (pattern III) – Hamstrings and psoas contraction
walk and develop a fixed contracture that will need surgical addressing, becoming itself a primary joint problem (Miller 2004). During adolescence there is a rapid decrease in the strength ratio (Miller 2004). The association of this factor with a spasticity growth impairs normal muscle development, leaving the young adults even weaker. In adults, muscle contractures due to spasticity develop much more rapidly than in childhood and results in severe amplitude limitations. The following fixed contractures are not associated with joint structural alterations, but they reduce the variability of movements available to motor control, promoting its decrease to a more gross pattern (Miller 2004). Nevertheless, joint deformities occur in adults with CP independently of the ambulatory status, diagnosis, or severity of involvement (Kembhavi et al. 2011). The combined impairments contribute to a sense of frustration in the adolescents, who lose the motivation to walk (Miller 2004). Adolescents with CP are usually heavier, and as such they are more prone to fall and be hurt during the adolescent clumsy stage (a normal feature of gait development). These falls may cognitively repress the gait behavior in this population, worsening the functional prognosis. However, it is difficult to assess if deterioration of mobility is due to the pathophysiology associated with CP, or the physical effects associated with aging, or an interaction of both (Kembhavi et al. 2011).
Spasticity Effect in Cerebral Palsy Gait
1043
Height and weight growth contributes to the development of the crouch gait. The muscles and joints become unable to support the body in the child typical toe-walking pattern, and knee flexion, hip flexion, and dorsiflexion increases, which can cause a midfoot and hindfoot collapse and a consequential severe planovalgus (Miller 2004). Young adults may also develop a back-kneeing pattern, particularly those using walking aids, with severe weakness of the gastrocnemius and often those who had tendon Achilles transections (Miller 2004).
Methodologies to Assess Spasticity During Gait in Cerebral Palsy Clinical evaluation of spasticity and its contribution to gait alterations in PC was traditionally a subjective analysis. Manual testing of spasticity uses non standardized velocities, which are frequently under the threshold for eliciting the stretch reflex, and is dependent on a relaxed state of the individual. As such it cannot differentiate between the contributions of neural and nonneural factors to spasticity. This is of relevance in the selection of treatment options, with medication used primarily in the first situation and casts or orthotics in the second. Also selective dorsal rhizotomy produces results only in a neural predominant spasticity (Bar-On et al. 2015). Different subjective scales have been used for clinical tone assessment, such as (i) the Ashworth Scale (AS)/Modified Ashworth scales (MAS) (Table X), (ii) the Tardieu Scale (TS) and the Modified Tardieu Scale (MTS) (Table Y), (iii) the Dynamic Evaluation Range of Motion (DAROM), and (iv) the Pendulum Test. However, as stated before, the reliability of these instruments have been criticized because the results obtained are dependent on some differences across inter-raters or day-testing conditions, e.g., the range of motion (ROM), the velocity of movement, and the position of the tested muscle (Boyd and Graham 1999). In specific, the Tardieu Scale and the Dynamic Evaluation of Range of Movement (DAROM) have reported more specifications that improved reliability in assessing spasticity, as, for example, the ROM is defined as slow or fast passive stretching. Moreover, the DAROM identifies a “range of motion deficit,” which means a value from the minimal muscle stretch position (Pandyan et al. 1999). When using the Ashworth Scales, an assessor tests the resistance to passive movement about a joint. Then, the resistance perceived while moving a joint through its full range of movement – except in grade “4” (Table 3) – is graded in a 5-Likert scale. The Ashworth Scale and the Modified Ashworth Scale can be used as measures of resistance to passive movement, but not as an ordinal level measure of spasticity (Pandyan et al. 1999; Damiano et al. 2002). The Tardieu Scale (Table 4) rates the spasticity as the difference between the reactions to stretch at two extreme velocities (Gracies et al. 2005): – The slowest: below the threshold of any significant stretch reflex (output = passive range of motion). – The fastest: maximizes the involvement of the stretch reflex, and if any spasticity is present, then the rater feels the sensations of catch and release/clonus/fatigable.
1044
M.C.N. Rosa and A.G.G. Roque
Table 3 The Ashworth and modified Ashworth scales Score 0 1
Ashworth scale No increase in tone Slight increase in tone giving a catch when the limb was moved in flexion or extension
1+
Slight increase in tone giving a catch when the limb was moved in flexion or extension
2
More marked increase in tone but limb easily flexed
3
Considerable increase in tone, passive movement difficult Limb rigid in flexion or extension
4
Modified Ashworth scale No increase in muscle tone Slight increase in muscle tone, manifested by a catch and release or by minimal resistance at the end of the range of motion when the affected part(s) is moved in flexion or extension Slight increase in muscle tone, manifested by a catch, followed by minimal resistance throughout the remainder (less than half) of the range of movement (ROM) More marked increase in muscle tone through most of the ROM, but affected part(s) easily moved Considerable increase in muscle tone, passive movement difficult Affected part(s) rigid in flexion or extension
Table 4 The Tardieu Scale principles and grading system Principles 1. Grading always performed: Muscle at rest before the stretch maneuver Reproducible velocity of stretch: once the fast velocity is selected for a muscle, it remains always the same At the same time of the day 2. Velocity of stretch Slow: V1, as slow as possible/slower than the rate of natural drop of the limb segment under gravity Fast: either V2 and V3; V2 is equivalent to the speed of the limb segment falling under gravity; V3 is equivalent to faster than the rate of natural drop of the limb segment under gravity
Grading X = spasticity angle (threshold) (Angle of arrest at slow speed (V1) – angle of catch at fast speed (V3)) Y = spasticity grade 0 = no resistance throughout passive movement 1 = slight resistance throughout passive movement 2 = clear catch at precise angle, interrupting passive movement, followed by release 3 = fatigable clonus (10s when maintaining pressure occurring at a precise angle) Catch without release Catch with minimal release Angle 0 For grades 0 and 1, spasticity angle X = 0 by definition
Comparing the characteristics of Ashworth and Tardieu Scales, it can be concluded that spasticity measured by the AS is confounded by muscle contracture, while the TS is able to differentiate spasticity from contracture and therefore is a more valid tool (Emily Patrick and Ada 2006). The Dynamic Evaluation of Range of Motion (DAROM) and the Pendulum Test complete the most widely used instruments for assessing spasticity in children with CP. The DAROM relies on the same velocity principle of TS, as it considers at least two different velocities of passive muscle stretching. The DAROM identifies a
Spasticity Effect in Cerebral Palsy Gait
1045
Table 5 Dynamic Evaluation of Range of Motion (DAROM) and pendulum tests outcomes Measurement Range of motion deficit (DROM) for DAROM tests: T1, T2, T3, T4
Parameter DROM I DROM II ASO
Pendulum Test
Ex.
t n
Description Range of motion deficit following a slow velocity stretch (V1) – expressed in degrees Range of motion deficit after a fast velocity stretch (V3) – expressed in degrees Value calculated as the difference between the DROM II and DROM I – expressed in degrees First swing excursion – difference between the starting angle and the first angle of reversal of the swinging limb; expressed in degrees Duration of the pendulum swings (sec.) Counting the maxima of the sinusoidal waves produced by the swinging limb after the heel was released
“range of motion deficit” (DROM), defined as a value from the minimal muscle stretch position. Using this instrument, two joint angles are measured: DROM I, defined as the range of motion deficit following a slow velocity stretch, and DROM II, defined as the angle of catch after a fast velocity stretch. The difference between DROM II and DROM I indicates the examined muscle group’s level of contracture and is called the angle of spasticity (ASO) (Bax et al. 2005). Otherwise, the Pendulum Test was first described by Wartenberg and it consists of a biomechanical method that evaluates muscle tone (discriminating between various degrees of spasticity) using gravity to induce the muscle stretch reflex during passive swinging of the lower leg. The following table (Table 5) shows a summary of the outcome measures that can be collected using the DAROM and Pendulum Tests (Domagalska et al. 2013). When considering how to treat spasticity, in addition to documenting changes in the resistance to passive movement (Pandyan et al. 1999), it may also be relevant to document changes in function. To further understand the role of spasticity in functional status of patients with CP, the literature has been focused in exploring possible relationships between spasticity and other gait parameters. Considering this problematic, there are still relevant controversies that may affect the quality of spasticity management in gait. For example, while Damiano and Abel (Abel et al. 2003) reported a significant correlation between the knee extensor Ashworth score and gait impairments, Gage (2004) and Boyd and Graham (1999) confirmed that rectus femoris spasticity is a possible cause of stiff knee gait; however, Wren et al. (2007) questioned the correlations between gait deviations and knee flexors spasticity and contractures. Moreover, spasticity in the knee (both in quadriceps and hamstrings) and hip muscles (adductors) (Ross and Engsberg 2007) has also been related with gait velocity and stride length decreasing (Damiano et al. 2006). This finding corroborates the clinical importance of spasticity in general functioning, as gait velocity is a major determinant of patients’ integration in community, i.e., someone that walks faster can easily and safety cross a street and therefore would not avoid community environments (Middleton et al. 2015). Spasticity and co-contraction have been critically assessed as being the same phenomenon (Diane et al. 2000). Despite the co-contraction is most of the times
1046
M.C.N. Rosa and A.G.G. Roque
related with spasticity and therefore simultaneously present in CNS disorders, these are not exactly the same phenomenon. In fact, co-contraction becomes excessive, and the agonist force decreases because spasticity transforms, at long term, the peripheral muscle components (configuration and properties of muscle fibers) and, consequently, generates abnormalities in dynamic muscle activation patterns. The application of EMG has been tested as a novel approach to quantitative study of muscle activity during dynamic motor activities. For example, during gait, the CP children stretching of calf muscles (agonist) in the post-contact stride period is known to be frequently accompanied by synchronous bursts in triceps surae (antagonist), generating abnormal muscle co-contraction patterns. This abnormal muscle patterns tend to be congruent with mechanical changes in ankle joint (e.g., absence of dorsiflexion). Even though EMG is able to detect these abnormal muscle patterns, it cannot reliably separate two possible causes for this phenomenon: the augmented myotatic reflex effect or the peripheral component contributions (muscle rigidity) (Crenna 1998). Instrumented gait analysis (IGA) has the unique ability to measure and provide information critical to the distinction between complex gait patterns. By assessing the tridimensional position of the different segments, including pelvis, at any moment of the gait cycle, it facilitates pattern recognition and problem addressing (Miller 2004). The integration of video analysis with dynamic electromyography (D-EMG) allows the identification of the contribution of muscles in each phase of the gait cycle, enhancing understanding of the pathological gait patterns (Miller 2004). Also, it makes it possible to assess the tridimensional gait results of the different proposed interventions (Miller 2004). By also incorporating clinical measures, IGA is the gold standard in gait evaluation in the CP population, with none other video analysis protocol reaching its consistency (Rathinam et al. 2014). Therefore, the simultaneous analysis of IGA and D-EMG during gait does not provide a direct information on the spasticity effect in locomotion patterns in children with cerebral palsy but can reliably assess consequences of spasticity in segment postures and spatiotemporal parameters. As a conclusion, the original definition of spasticity (Lance and McLeod 1981) “does not include impaired movement and an abnormal posture” and only considered observations under static conditions (muscles relaxed; muscles tonically activated), which explains the poor report of correlations between the pathological responsiveness to stretch measured at rest and motor deficits assessed during natural actions, such as gait (Crenna 1998). Therefore, the lack of adequate instruments to assess spasticity during gait has been limiting the consensus about the effect of spasticity in gait pattern.
Spasticity Treatments with Impact in CP Gait For some years now, a spinal cord mechanism that controls the excitability of the stretch reflex has been described, which does not depend on supraspinal inhibition. The so-called postactivation depression is an intrinsic property of the Ia afferent fibers that is associated with a decreased release of neurotransmitters and has been
Spasticity Effect in Cerebral Palsy Gait
1047
shown to be reduced with limb immobilization in a shortened position. As such, muscle atrophy and reduced activity seen in CP also results in spasticity. More importantly this mechanism seems to be affected by active movement, passive mobilization, and prolonged muscle stretching, all of which can prevent muscle hypertonia (Trompetto et al. 2014). This could be one mechanism subjacent to improvements seen with both splits/orthotics and physiotherapy movement modalities. Bracing and orthotics are classic treatment options for lower limb muscle and tendon lengthening in CP. Orthotics applied in the ankle, holding the foot in a plantigrade position and correcting planovalgus foot deformities, may provide the stable fundamental base that allow CP children to develop balance and train taskspecific activities in stance but also give a movement focus in the knee and hip. This favors a more symmetric and “normal” gait pattern. However, by restraining active plantar flexion, ankle-foot orthotics (AFO) reduce the lift-off moment of gait, producing secondary adaptations, such as the hip extensors being the main responsible for the push-off phase and forward movement, which is frequently combined with pelvis rotation. A flexible leaf-spring AFO may allow some degree of plantar flexion but only maintains the foot in dorsiflexion if there is absent spasticity of the gastrocnemius and soleus. Baclofen is a gamma-aminobutyric acid (GABA) agonist and thus mimics an inhibitory neurotransmitter, reducing spasticity. It has been used both orally and, more recently, intrathecally. The latter reduces some side effects such as excessive sedation and also increases the bioavailability of the drug in its action sites: the cerebrospinal fluid and brain. Its evidence is insufficient in regard to decreasing spasticity, improving gross motor actions, and transferring achievements to activities and participation (Martin et al. 2010). Despite the lack of evidence, baclofen is a treatment option for generalized spasticity, such as the one presented in tetraparetic CP. Other alternatives to the management of generalized spasticity are pharmacological, including diazepam and tizanidine with moderate evidence and dantrolene with insufficient evidence. However, the pharmacological options have not demonstrated changes in gross motor function or quality of life (Martin et al. 2010). Botulinum toxin A (BTA) is a neurotoxin that inhibits acetylcholine release into the neuromuscular junction, reducing the muscle response to efferent neural stimuli. As such, the tonic and phasic stretch reflexes and consequently spasticity are diminished in the selected injected muscles or muscular groups. However, the duration of its effect is estimated in 3–6 months and the action potential of the muscle and active movements are reduced, which means that BTA injections by themselves are not a treatment for spasticity-derived gait alterations but can provide the means to facilitate a more biomechanical correct movement and posture during gait (Klemens Fheodoroff et al. 2016). This facilitation is achieved by providing the correct afferent exteroceptive and proprioceptive stimulus in an intensive physiotherapy program or the use of orthotics. Its direct impact is therefore in the structural and functional dimensions of the International Classification of Functioning, Disability and Health (ICF), with only a latter translation to activities and participation,
1048
M.C.N. Rosa and A.G.G. Roque
due to the time required to motor relearning of the treated limb to occur (Love et al. 2010). Precocious use of BTA in young children with CP may reduce the evolution to a crouch gait and the need for surgical corrective procedures but also increase the probability of the child to achieve the motor milestones crucial to a normal neurodevelopment (Klemens Fheodoroff et al. 2016). Furthermore it is considered a safe procedure in CP children with localized spasticity (Delgado et al. 2010). An international consensus has stated in 2010 that BTA is effective in the management of spastic equines to improve gait (level A), however, only probably effective to improve goal attainment and function in the management of spastic equinus (level B). Also, adductor injections may help attain some goals (level B) and may delay hip displacement (level A) but does not improve gross motor function and does not affect long-term outcome (level A). Multiple lower limb injections have conflicting information as respect to gait, goal attainment, and function (level U) (Love et al. 2010). Another treatment option for spasticity-related gait disorders in PC is selective dorsal rhizotomy (SDR). By selectively impairing afferent stimuli to the medullae, this procedure inhibits the efferent response of the stretch reflex, resulting in decreased tone and therefore reducing spasticity. Since the clinical effects are somewhat similar to BTA or baclofen, it is expected to act merely in the structural and functional dimensions of ICF, with the considerations made to BTA regarding transfer to activities and participation, being also valid. However, the benefits observed in the sagittal plane at all lower limb joints with SDR are associated with an augmented misalignment in the pelvis. These postural alterations include crouching and horizontalization of the sacrum (Roberts et al. 2015), which will progress to the adult age due to the irreversibility of the technique. Furthermore, the procedure may reveal underlying muscle weakness that worsens the performance of previous ambulatory children, especially if the antigravity postural capacity is affected. Thus, a multidimensional evaluation and biomechanical analysis of gait should be performed to guide patients’ election to this procedure (Grunt et al. 2014). Indeed, spasticity of the extensor apparatus of the lower limb may be a useful characteristic of CP gait, as it maintains the possibility of placing the foot (even if in equinus) on the ground and thus relieve the body mass loading in the contralateral limb. This is a fundamental characteristic of a normal gait. When spasticity is removed, for instance, by the abovementioned dorsal rhizotomy, this extension maintenance is impaired, resulting in incapacity of body mass transfer to the affected limb. The correct consideration of the contribution of spasticity to CP gait is crucial to the success or failure of a spasticity management procedure, when it relates to activity, participation, and transfer to the child environment (Grunt et al. 2014). Despite the absent role in the management of tone abnormalities of physiotherapy, it is a fundamental adjunctive therapy to increase muscle strength and aerobic conditioning but also in improving lower extremity coordination and speed (Martin et al. 2010). Excessive exercise and strength training was said to aggravate spasticity in CP and therefore a cause to gait deterioration; however, studies have challenged this traditional belief (DL 2006). As such intensive task-oriented practice is nowadays a central practice in CP rehabilitation, which in the case of gait precludes
Spasticity Effect in Cerebral Palsy Gait
1049
intensive treadmill training with or without assistive weight supports, despite a lack of demonstrated evidence (Martin et al. 2010). Other physiotherapy modalities that address spasticity and gait have long been used in clinical practice. Neurodevelopmental therapy (NDT) and functional training (FT) evaluate the individual as a whole, directing its practice according to the identified incapacities, limitations, or handicaps (Martin et al. 2010). By focusing in the particular characteristics of the different presentations of gait in CP instead of intervening in a cluster of typical patterns, these modalities have struggled to produce conclusive evidence in investigational studies designed as randomized control trials (RCT) but have nevertheless a role in neurorehabilitation of the consequences of spasticity in CP gait, either individually or as multimodal approaches. Virtual reality (VR) is an emerging computer-assisted multisensorial feedback approach to neurorehabilitation that promotes neuroplasticity induced by repetition in an enjoyable environment but depends on the capacity of the individual to integrate the “scene.” By conferring the possibility of manipulation of the environment, VR facilitates motor learning which in the case of the present chapter is directed to improvements in the spatiotemporal features of CP gait (E. Monge Pereiraa et al. 2014). Transcranial magnetic stimulation (TMS) has also recently been therapeutically used, and it is characterized by noninvasive stimulation of the motor cortex, promoting supraspinal inhibition of spasticity and therefore reducing CP spasticity-induced gait changes (Gunduz et al. 2014). When not efficiently managed, spasticity can progress to the complications previously mentioned, and orthopedic surgery has a pivotal role in addressing those consequences. As deformity evolves from dynamic to fixed, the focal tone management modalities already referred become contraindicated, as bone deformity or soft tissue contractures are not tone dependent and therefore susceptible. A wide range of orthopedic interventions are available in CP management, some of which address soft tissues such as tendon lengthening and ligament or capsular releases, while others intervene in the skeletal system with osteotomies and arthrodeses as the mainstay of the possibilities. After a biomechanical evaluation, there is also the possibility of combining surgical procedures to different anatomical locations, in order to optimize gait, being referred to as single-event multilevel surgery (SEMS) (Thomason et al. 2013). We previously described the pattern of flexed knee gait, characterized by the incapacity to perform full hip and knee extension in late stance. The surgical lengthening of the biarticular gastrocnemius, hamstrings, and iliopsoas is one component of surgical procedures targeting improved stance and gait in spastic diplegia. Indeed, a surgical procedure that lengthens the gastrocnemius may overcome the hip secondary compensation abovementioned, as it allows some active plantar flexion and also reduces the contracture in midstance, resulting in global gait improvements. Also hamstring lengthening is effective in correcting knee extension; however, since the hamstrings contribute to hip extension in stance, the procedure affects to some degree pelvic control. Other surgical options have been developed, such as the supracondylar extension osteotomy to correct the flexed knee deformity combined with patellar tendon advancement, which is described to improve knee extensor lag,
1050
M.C.N. Rosa and A.G.G. Roque
knee extension in gait, and overall gait. Due to increased stability in the knee, these procedures are also associated with reduced spasticity in the knee extensors (Sossai et al. 2010). Surgical selection should thus be decided based in gait analysis of the CP individual potentiating the after gains. From the treatments referred above, orthopedic surgery (including tendon transfer, muscle lengthening, or others), orthotics, and spasticity control (including rhizotomy, BTA, phenol blocks, or others) have succeeded in improving gait velocity, decreasing cadence, and increasing stride length (Paul et al. 2007). Other treatments such as serial casting, muscle strengthening exercises, and biofeedback have not accumulated the necessary evidence to affirm their effects in improving CP spatiotemporal components of gait. A multifactorial approach to gait alterations should thus be considered in detrimental of more isolated approaches.
Future Directions There is a scarce of consolidated information on the evolution of the gait pattern throughout adolescent and adulthood, specially from CP individuals who are subjected to orthopedic and invasive tone management procedures (Wilson et al. 2014). This is an issue that must be address in future studies, as the best medical, developmental, and rehabilitative science for optimizing mobility, motor functioning, and fitness for adults with CP are unknown (Frisch and Msall 2013). Also, it is not clear for now what the activity limitations that impair adults with CP from participating in their community are and also which environmental factors and in what measure they affect CP adults’ handicaps, despite some works investigating this theme (Frisch and Msall 2013). Given the difficulty to distinguish between spasticity and intrinsic hypertonia with the use of current evaluation measures, question remains if the actual treatments for lower limb hypertonia really address spasticity or are actually just influencing the biomechanical factors responsible for the intrinsic hypertonia as a result of immobilization. This is an important issue when deciding for intervention use, orthopedic or tone management. In other words where and when are the limits to tone management potential? When does rigidity become a primary issue, contraindicating spasticity control? The use of biomechanical measures with EMG recordings and perhaps accelerometers should be included in future clinical trials of CP gait evaluation and spasticity management, in order to differentiate between the focus of the treatment and provide a more comprehensive knowledge of the individual alterations (Chang et al. 2010), and on where in the gait cycle the velocity threshold to elicit the tonic or phasic stretch reflex occurs.
References Abel MF, Damiano DL, Blanco JS, Conaway M, Miller F, Dabney K, Sutherland D, Chambers H, Dias L, Sarwark J, Killian J, Doyle S, Root L, LaPlaza J, Widmann R, Snyder B (2003) Relationships among musculoskeletal impairments and functional health status in ambulatory cerebral palsy. J Pediatr Orthop 23(4):535–541
Spasticity Effect in Cerebral Palsy Gait
1051
Awaad Y, Rizk T (2012) Spasticity in children. J Taibah Univ Med Sci 7(2):53–60 Bar-On L, Molenaers G, Aertbelien E, Van Campenhout A, Feys H, Nuttin B, Desloovere K (2015) Spasticity and its contribution to hypertonia in cerebral palsy. BioMed Res Int 2015:1–10 Bax M, Goldstein M, Rosenbaum P, Leviton A, Paneth N, Dan B, Jacobsson B, Damiano D (2005) Proposed definition and classification of cerebral palsy. Dev Med Child Neurol 47:571–576 Boyd R, Graham HK (1999) Objective measurement of clinical findings in the use of botulinum toxin type A for the management of children with CP. Eur J Neurol 6:23–35 Chang FM, Rhodes JT, Flynn KM, Carollo JJ (2010) The role of gait analysis in treating gait abnormalities in cerebral palsy. Orthop Clin N Am 41(4):489–506 Crenna P (1998) Spasticity and ‘Spastic’ gait in children with cerebral palsy. Neurosci Biobehav Rev 22(4):571–578 Damiano DL (2006) Activity, activity, activity: rethinking our physical therapy approach to cerebral palsy. Phys Ther 86(11):1534–1540 Damiano DL, Quinlivan JM, Owen BF, Payne P, Nelson KC, Abel MF (2002) What does the Ashworth scale really measure and are instrumented measures more valid and precise? Dev Med Child Neurol 44(2):112–118 Damiano DL, Laws E, Carmines DV, Abel MF (2006) Relationship of spasticity to knee angular velocity and motion during gait in cerebral palsy. Gait Posture 23(1):1–8 Dana R, Cub DA (2013) Cerebral palsy gait, clinical importance. MAEDICA – J Clin Med 8(4):388–393 de Morais Filho MC, Kawamura CM, Lopes JAF, Neves DL, Cardoso M d O, Caiafa JB (2014) Most frequent gait patterns in diplegic spastic cerebral palsy. Acta Ortop Bras 22:197–201 Delgado MR, Hirtz D, Aisen M, Ashwal S, Fehlings DL, McLaughlin J, Morrison LA, Shrader MW, Tilton A, Vargus-Adams J (2010) Practice parameter: pharmacologic treatment of spasticity in children and adolescents with cerebral palsy (an evidence-based review): report of the quality standards subcommittee of the American academy of neurology and the practice committee of the child neurology society. Neurology 74(4):336–343 Diane L, Martellotta L, Sullivan J, Granata P, et al. (2000) Muscle force production and functional performance in spastic cerebral palsy: Relationship of cocontraction. Archives of Physical Medicine and Rehabilitation 81(7):895–900 Domagalska M, Szopa A, Syczewska M, Pietraszek S, Kidoń Z, Onik G (2013) The relationship between clinical measurements and gait analysis data in children with cerebral palsy. Gait Posture 38(4):1038–1043 Fheodoroff K, Jacinto J, Geurts A, Molteni F, Franco JH, Santiago T, Rosale R, Gracies J-M (2016) How can we improve current practice in spastic paresis? Eur Neurol Rev 11(2) (Epub ahead of print) Filloux FM (1996) Neuropathophysiology of movement disorders in cerebral palsy. J Child Neurol 11(Suppl 1):S5–S12 Frisch D, Msall ME (2013) Health, functioning, and participation of adolescents and adults with cerebral palsy: a review of outcomes research. Dev Disabil Res Rev 18(1):84–94 Gage JR (2004) The treatment of gait problems in cerebral palsy. Mac Keith, London Gracies J-M, Burke K, Clegg NJ, Browne R, Rushing C, Fehlings D, Matthews D, Tilton A, Delgado MR (2005) Reliability of the Tardieu Scale for assessing spasticity in children with cerebral palsy. Arch Phys Med Rehabil 91(3):421–428 Grunt S, Fieggen AG, Vermeulen RJ, Becher JG, Langerak NG (2014) Selection criteria for selective dorsal rhizotomy in children with spastic cerebral palsy: a systematic review of the literature. Dev Med Child Neurol 56(4):302–312 Gunduz A, Kumru H, Pascual-Leone A (2014) Outcomes in spasticity after repetitive transcranial magnetic and transcranial direct current stimulations. Neural Regen Res 9(7):712–718 Hobart JC, Riazi A, Thompson AJ (1994) Spasticity: a review. Neurology 44(11):S12–S20 Kembhavi G, Johanna D, Kelsey P, Devon P (2011) Adults with a diagnosis of cerebral palsy: a mapping review of long-term - Review Development medicine and child neurology 53(7):610–614 Lance JW (1980) Symposium synopsis. In: Feldman RG, Young RR, Koella WP (eds) Spasticity: disordered motor control. Chicago, Yearbook Medical Publishers
1052
M.C.N. Rosa and A.G.G. Roque
Lance JW, McLeod JG (1981) A physiological approach to clinical neurology. Butterworths, London Love SC, Novak I, Kentish M, Desloovere K, Heinen F, Molenaers G, O’Flaherty S, Graham HK (2010) Botulinum toxin assessment, intervention and after-care for lower limb spasticity in children with cerebral palsy: international consensus statement. Eur J Neurol 17:9–37 Martin L, Baker R, Harvey A (2010) A systematic review of common physiotherapy interventions in school-aged children with cerebral palsy. Phys Occup Ther Pediatr 30(4):294–312 Middleton A, Fritz SL, Lusardi M (2015) Walking speed: the functional vital sign. J Aging Phys Act 23(2):314–322 Miller F (2004) Cerebral palsy. Springer, New York Monge Pereiraa E, Rueda FM, Diego IMA, de la Cuerda RC, de Mauroc A, Miangolarra JC (2014) Empleo de sistemas de realidad virtual como método de propiocepción en parálisis cerebral: guía de práctica clínica. Neurologia 29(9):550–559 Novacheck TF, Stout JL, Tervo R (2000) Reliability and validity of the Gillette Functional Assessment Questionnaire as an outcome measure in children with walking disabilities. J Pediatr Orthop 20(1):75–81 O’Byrne JM, Jenkinson A, O’Brien TM (1998) Quantitative analysis and classification of gait patterns in cerebral palsy using a three-dimensional motion analyzer. J Child Neurol 13(3):101–108 Pandyan AD, Johnson GR, Price CIM, Curless RH, Barnes MP, Rodgers H (1999) A review of the properties and limitations of the Ashworth and modified Ashworth Scales as measures of spasticity. Clin Rehabil 13(5):373–383 Patrick E, Ada L (2006) The Tardieu Scale differentiates contracture from spasticity whereas the Ashworth Scale is confounded by it. Clin Rehabil 20(2):173–182 Paul SM, Siegel KL, Malley J, Jaeger RJ (2007) Evaluating interventions to improve gait in cerebral palsy: a meta-analysis of spatiotemporal measures. Dev Med Child Neurol 49(7):542–549 Piccinini L, Cimolin V, Galli M, Berti M, Crivellini M, Turconi AC (2007) Quantification of energy expenditure during gait in children affected by cerebral palsy. Eura Medicophys 43(1):7–12 Rathinam C, Bateman A, Peirson J, Skinner J (2014) Observational gait assessment tools in paediatrics – a systematic review. Gait Posture 40(2):279–285 Roberts A, Stewart C, Freeman R (2015) Gait analysis to guide a selective dorsal rhizotomy program. Gait Posture 42(1):16–22 Rodda J, Graham HK (2001) Classification of gait patterns in spastic hemiplegia and spastic diplegia: a basis for a management algorithm. Eur J Neurol 8:98–108 Rodda JM, Graham HK, Carson L, Galea MP, Wolfe R (2004) Sagittal gait patterns in spastic diplegia. J Bone Joint Surg Br 86(2):251–258 Ross SA, Engsberg JR (2007) Relationships between spasticity, strength, gait, and the GMFM-66 in persons with spastic diplegia cerebral palsy. Arch Phys Med Rehab 88(9): 1114–1120 Scholtes VAB, Becher JG, Beelen A, Lankhorst GJ (2006) Clinical assessment of spasticity in children with cerebral palsy: a critical review of available instruments. Dev Med Child Neurol 48(1):64–73 Sossai R, Vavken P, Brunner R, Camathias C, Graham HK, Rutz E (2010) Patellar tendon shortening for flexed knee gait in spastic diplegia. Gait Posture 41(2):658–665 Thomason PJ, Selber P, Graham HK (2013) Single event multilevel surgery in children with spastic cerebral palsy: 5 year prospective cohort study. Gait and Posture 37(1):23–28 Trompetto C, Marinelli L, Mori L, Pelosin E, Currà A, Molfetta L, Abbruzzese G (2014) Pathophysiology of spasticity: implications for neurorehabilitation. BioMed Res Int 2014:1–8 Wilson NC, Chong J, Mackey AH, Stott NS (2014) Reported outcomes of lower limb orthopaedic surgery in children and adolescents with cerebral palsy: a mapping review. Dev Med Child Neurol 56(9):808–814 Wren TA, Do KP, Hara R, Dorey FJ, Kay RM, Otsuka NY (2007) Gillette gait Index as a gait analysis summary measure: comparison with qualitative visual assessments of overall gait. J Pediatr Orthop 27(7):765–768
Natural History of Cerebral Palsy and Outcome Assessment Erich Rutz and Pam Thomason
Abstract
Cerebral palsy (CP) is defined as “a group of permanent disorders of the development of movement and posture, causing activity limitation, that are attributed to nonprogressive disturbances that occurred in the developing fetal or infant brain. The motor disorders of CP are often accompanied by disturbances of sensation, perception, cognition, communication, and behavior, by epilepsy; and by secondary musculoskeletal problems.” Though the neurological disturbance is nonprogressive, the natural history is for deterioration in gait and motor function due to musculoskeletal pathology. The natural history and outcomes of musculoskeletal problems are discussed. The measurement tools and comprehensive assessment used to evaluate gross motor function and gait impairments in children with CP are presented. Keywords
Cerebral palsy • Musculoskeletal deformity • Assessment • Gait • Hip displacement • Outcome
Contents State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Natural History, Spasticity, Muscle Force, and Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Hip and Spine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Orthopedic Surgery to Improve Gait and Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1054 1054 1056 1058 1059
E. Rutz (*) Pediatric Orthopaedic Department, University Children’s Hospital Basel, Basel, Switzerland e-mail: [email protected] P. Thomason Hugh Williamson Gait Analysis Laboratory, Royal Children’s Hospital, Melbourne, VIC, Australia e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_49
1053
1054
E. Rutz and P. Thomason
Outcome Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Body Structure and Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Functional Mobility Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Functional Assessment Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1061 1063 1066 1066 1066 1068 1068 1069
State of the Art Natural History, Spasticity, Muscle Force, and Growth Natural history of gait in children with cerebral palsy (CP) is one of deterioration. Several studies report on these findings (Norlin and Odenrick 1986; Johnson et al. 1997; Bell et al. 2002; Rose et al. 2010; Morgan and McGinley 2014). The most important factors that affect the natural history of deterioration, in our opinion, are the following: increased muscle tone during growth, the development of deformities, and a decrease of muscular force. Primary features in children with CP are spasticity, muscle weakness, and lack of selective motor control. All these factors are in conflict with growth in children with CP. Lower limbs deformities are worsened by the increase of body weight (Davids et al. 2015). A longitudinal study (Johnson et al. 1997) over a mean of 32 months on 18 subjects with spastic diplegia, ranging in age from 4 to 14 years was conducted. Instrumented gait analyses (IGA) were performed to compare the temporal and kinematic data across the two time intervals. The comparison revealed a deterioration of gait stability, evidenced by increases in double support and decreases in single support time with time and growth. In this study, kinematic analysis revealed a loss of excursion about the knee, ankle, and pelvis. Additionally, passive range-ofmotion analysis revealed a decrease in the popliteal angle over time. They concluded that, in contrast to the gait of children with intact motor function, ambulatory ability tends to worsen over time in spastic CP (Johnson et al. 1997). Bell et al. (2002) showed a deterioration of gait temporal and stride parameters and kinematics over time in 28 children with CP. There was no surgical intervention in this cohort during the study period of 4.4 years. Of the 28, 19 children were classified with the diagnosis of diplegia, seven with hemiplegia, and two with quadriplegia. In general, CP is considered a static deformity. But in this study (Bell et al. 2002), changes in function commonly result with growth and age. At the time of the first 3DGA, recommendations for 14 were conservative (e.g., bracing, botulinum toxin A). After 4.4 years, at the end of the study, surgery was recommended in all but two children. Hagglund and Wagner, (2008) reported the development of spasticity with age in children with CP. In 1994, a register and a health-care program for children with CP in southern Sweden were initiated. In the program, the child’s muscle tone according to the Modified Ashworth Scale is measured twice a year until 6 years of
Natural History of Cerebral Palsy and Outcome Assessment
1055
age and then once a year. They used this data to analyze the development of spasticity with age in a total population of children with CP. In their study, all measurements of muscle tone in the gastrocnemius-soleus muscle in all children with CP from 0 to 15 years during the period 1995–2006 were analyzed. They found that in the total sample, the degree of muscle tone increased up to 4 years of age. After 4 years of age, the muscle tone decreased each year up to 12 years of age. A similar development was seen when excluding the children operated with selective dorsal rhizotomy, intrathecal baclofen pump, or tendo achilles lengthening. At 4 years of age, 47% of the children had spasticity in their gastro-soleus muscle graded as Ashworth 2–4. After 12 years of age, 23% of the children had that level of spasticity. Hagglund and Wagner, (2008) concluded that, in children with CP, the muscle tone as measured with the Modified Ashworth Scale increases up to 4 years of age and then decreases up to 12 years of age. Morgan and Mc Ginley conducted a systematic review on gait function and decline in adults with CP. Sixteen studies met the inclusion criteria, and the reported mean ages of the study groups varied from 22 to 42.6 years of age. Results suggest that mobility decline occurs in 25% or more of adults with CP. Those at higher risk of gait decline are those with worse initial gait ability, bilateral rather than unilateral motor impairment, older age, and higher levels of fatigue and pain. They conclude that many ambulant adults with CP experience mobility decline earlier than their nondisabled peers (Morgan and McGinley 2014). The relationship of strength, weight, age, and function in ambulatory children with CP was investigated by Davids et al. (2015). The study design was prospective, case series of 255 subjects, aged 8–19 years, with diplegic type of CP. Linear regression was used to predict the rate of change in lower extremity muscle strength, body weight, and strength normalized to weight (STR-N) with age. The cohort was analyzed as a whole and in groups based on functional impairment as reflected by Gross Motor Function Classification System (GMFCS) level. In their results, they found that strength increased significantly over time for the entire cohort at a rate of 20.83 N/y (P = 0.01). Weight increased significantly over time for the entire cohort at a rate of 3.5 kg/y (P < 0.0001). Lower extremity STR-N decreased significantly over time for the entire cohort at a rate of 0.84 N/kg/y (P < 0.0001). The rate of decline in STR-N (N/kg/y) was comparable among age groups of the children in the study group. Interestingly they stated that there were no significant differences in the rate of decline of STR-N (N/kg/y) among GMFCS levels. There was a 90% chance of independent ambulation (GMFCS levels I and II) when STR-N was 21 N/kg (49% predicted relative to typically developing children). The results of this study (Davids et al. 2015) support the long-standing clinically based observation that STR-N decreases with age in children with CP. This decrease occurs throughout the growing years and across GMFCS levels I to III. Independent ambulation becomes less likely as STR-N decreases. The natural history of gross motor development in children with CP aged 1–15 years was reported by Beckung et al. (2007). These curves may be useful for monitoring and predicting motor development, for planning treatment, and for evaluating outcome after interventions.
1056
E. Rutz and P. Thomason
Fig. 1 Indicates the typical posture of a child with spastic diplegia
In all of these studies, the natural history of ambulation in children with CP consists of deterioration over time. (Norlin and Odenrick 1986; Bell et al. 2002; Rose et al. 2010; Rutz et al. 2012; Morgan and McGinley 2014; Davids et al. 2015; Graham et al. 2016). Figures 1 and 2 present illustrative postures of the lower extermities (1 in bilateral spastic CP, 2 in unilateral spastic CP).
The Hip and Spine Soo et al. (2006) reported that hip displacement is common in children with CP, with an overall incidence of 35% found in their study. The risk of hip displacement is directly related to gross motor function as graded with the GMFCS (Palisano et al. 1997). There are four studies from hip surveillance programs, and all report very similar results (Soo et al. 2006; Connelly et al. 2009; Terjesen 2012; Hagglund et al. 2014). Hips in ambulatory children with CP are usually stable or easily made so, but hips in children with GMFCS levels IV and V have unstable hips (Rutz 2012; Rutz et al. 2015). The untreated hip dislocation is not an easy problem. Pritchett (1983) reported the incidence of pain and other complications associated with the untreated spastic dislocated or subluxed hip in 80 institutionalized patients. The average age was 19 years, and the average length of follow-up study was 12 years. Eighty-five percent of the patients had scoliosis, and 56% had significant pelvic obliquity. A dislocated hip predisposed to lower extremity fractures but did not present problems of pain, decubitus ulcers, or difficulties with perineal hygiene. The loss of sitting balance, scoliosis, and pelvic obliquity are correlated with the severity of neurologic involvement rather than with the mechanics of a dislocated hip. Several studies reported pain and other problems as a significant finding (Samilson et al. 1972; Moreau et al. 1979). Knapp et al. (Knapp and Cortes 2002) reported seven dislocated hips (18%, out of 29 hips) were definitely painful and four hips (11%) produced only mild or intermittent pain. Twenty-seven hips (71%) were not painful. Pruszczynski and Miller (2015) performed a review and reported that when hip displacement in children with CP is identified early, treatment is more successful.
Natural History of Cerebral Palsy and Outcome Assessment
1057
Fig. 2 Presents a child with spastic hemiplegia on the right side
The review included ten studies with sample sizes greater than 20 children. They found that the GMFCS level has a strong impact on subluxation risk and that the risk continues to the end of growth. Wawrzuta et al. 2016 studied “hip health” in a population-based cohort of adolescents and young adults with CP to investigate associations between hip morphology, pain, and gross motor function. Ninety-eight young adults were reviewed at a mean age of 18 years, 10 months. Hip morphology was classified using the Melbourne Cerebral Palsy Hip Classification Scale (MCPHCS) (Murnaghan et al. 2010). Hip pain was reported in 72% of participants. Poor hip morphology at skeletal maturity was associated with high levels of pain. Limited hip surveillance and access to surgery, rather than GMFCS, were associated with poor hip morphology. The majority of young adults with access to hip surveillance and preventive and reconstructive surgery had satisfactory hip morphology at skeletal maturity and less pain. The role of pelvic obliquity is not entirely clear, but its management deserves closer scrutiny in children and adolescents with CP (Heidt et al. 2015). Natural history of scoliosis in spastic CP was described very well in the classical paper of Saito et al. (1998). They concluded that the risk factors for progression of scoliosis in spastic CP are having a spinal curve of 40 before age 15 years, having
1058
E. Rutz and P. Thomason
total body involvement, being bedridden, and having a thoracolumbar curve. Patients with these risk factors might benefit from early surgical intervention to prevent progression to severe scoliosis (Saito et al. 1998). Gu et al. (2011) performed a retrospective study to analyze the development and progression of scoliosis in children and adolescents with nonambulatory spastic quadriplegic CP. A total of 110 children and adolescents 40 by the age of 12 years were associated with greater increases in Cobb angle with age (Gu et al. 2011).
Orthopedic Surgery to Improve Gait and Function The current concept for the management of musculoskeletal deformities in children with CP is single-event multilevel surgery (SEMLS). In a systematic review of SEMLS evidence was found for large improvements in gait dysfunction, moderate improvements in health-related quality of life, and only small changes in gross motor function (McGinley et al. 2012). In the first randomized clinical trial of SEMLS, a 50% improvement in gait function and a 4.9% improvement in gross motor function (GMFM-66) were reported (Thomason et al. 2011). The 5-year results of this clinical trial show that these improvements were largely maintained at 5 years after SEMLS (Thomason et al. 2013). In the SEMLS approach, the gait pattern is identified and evaluated by IGA as part of the diagnostic matrix (Davids et al. 2003). SEMLS in bilateral spastic CP can be considered an exercise in correcting anatomical deformities based on the clinical findings. It is necessary to consider all components of the matrix so that surgical planning is optimized for the individual child. Careful preoperative discussions about setting realistic goals help to ensure that the patient’s, parents’, and surgeon’s goals are consistent and achievable (Thomason and Graham 2013). A comprehensive plan is then developed for the correction of all muscle tendon contractures, torsional malalignments, and joint instabilities in one operative session. Rehabilitation requires at least 1 year and improvements continue into the second year, postoperatively. A carefully tailored and carefully monitored rehabilitation program can ensure that the child will reach a higher level of function (Thomason et al. 2013; Thomason and Graham 2013). Follow-up until skeletal maturity is recommended, to detect for new or recurrent deformities. The orthotic prescription must be carefully monitored throughout the first year after surgery. Functional recovery and orthotic prescription can be monitored by a gait laboratory visit every 3 months for the first year after surgery and yearly thereafter. Our approach to SEMLS rehabilitation has been described in more detail elsewhere (Thomason and Graham 2013).
Natural History of Cerebral Palsy and Outcome Assessment
1059
Multilevel orthopedic surgery for older children and adolescents with severe crouch gait is effective for relieving stress on the knee extensor mechanism, reducing knee pain, and improving function and independence (Rodda et al. 2006). Surgical treatment for equinus gait in children with spastic diplegia was successful, at a mean of 7 years, in the majority of cases when combined with multilevel surgery, orthoses, and rehabilitation. No patient developed crouch gait, and the rate of revision surgery for recurrent equinus was 12.5% (Firth et al. 2013).
Classification Classification of functional levels in children with CP permits description of the level of impairment, encourages clear communication, and is important to establish gross motor prognosis, monitoring progress and in choice of appropriate interventions (Rosenbaum et al. 2002, 2007a). Development of the Gross Motor Functional Classification System (GMFCS) has given clinicians and parents a common language to communicate about CP (Palisano et al. 1997, 2008). The GMFCS is a classification system, and its use is essential when discussing gross motor function in children with CP. It should be used alongside the classification of upper limb function (Manual Ability Classification System) and communication abilities (Communication Function Classification System) to provide the essential context for considering the individual child’s prognosis, goal setting, management, and outcome evaluation (Eliasson et al. 2006; Hidecker et al. 2011). The descriptors and accompanying illustrations are found in Fig. 3a, b. Although GMFCS levels are considered to be stable over time, it is important to note that the descriptors for youth aged 13–18 years allow for deterioration in gross motor function, a decrease in independent walking abilities, and an increase in the need for both assistive devices and wheeled mobility. This is particularly true at GMFCS levels II, III, and IV (McCormick et al. 2007). Certain musculoskeletal features and deformities are closely related to GMFCS level. The shape of the proximal femur shows a strong correlation with GMFCS level. (Robin et al. 2008). The incidence and severity of hip displacement are directly predicted by GMFCS level (Soo et al. 2006). The relationship between GMFCS and hip displacement has implications for screening and management protocols. Changes in GMFCS levels should be carefully documented. The GMFCS may not be stable in the very young child. However, the most common reason for a change in GMFCS level is an error in interpretation in the previous or current examination. Given that the GMFCS is a categorical grading system, true changes in GMFCS level sometimes occur which may be in both directions, namely, improvement or deterioration. After major intervention such as selective dorsal rhizotomy (SDR) or single-event multilevel surgery (SEMLS), a few children move up a level. This is uncommon and should not be expected in more than 5–10% of children (Rutz et al. 2012). Deterioration in GMFCS level is more common. For example, lengthening of the Achilles’ tendons in children in GMFCS level II can result in progressive crouch gait and the need for assistive devices. For these children, their gait and function
Fig. 3 (a) GMFCS levels descriptors for ages 6–12 years (a) and 13–18 years (b)
1060 E. Rutz and P. Thomason
Natural History of Cerebral Palsy and Outcome Assessment
1061
deteriorate, and GMFCS level changes from II to III (Rodda et al. 2006). The relative stability of the GMFCS means that this is not an outcome measure and was never meant to be one. The GMFCS provides an excellent guide to long-term prognosis and influences our decision making and management plan. The development of gross motor function in children with CP can be described by a series of curves (Rosenbaum et al. 2002). Understanding the position of a child’s development in relation to their gross motor curve provides a rational basis for understanding management strategies, goal setting, and long-term gross motor function (Hanna et al. 2008). The popularity of many forms of intervention in early childhood in children with CP is the mistaken attribution of improvements in gross motor function to the intervention, when natural history has such a large effect. Association is not causation. In most children, gross motor function reaches a plateau between 3 and 6 years, with some regression in later childhood. One of the causes for this regression in gross motor function is progressive musculoskeletal pathology (Graham 2004). After age 6 years, gait parameters deteriorate as contractures and bony deformities increase. It should be noted that the gross motor curves (Rosenbaum et al. 2002) include GMFM data up to the age of 15 years. There is much less information regarding changes in gross motor function in the 15–20 age group and beyond. Changes in gross motor function and gait during the plateau/early decline phase can be more realistically attributed to intervention.
Outcome Assessment Previously we have discussed the natural history of the musculoskeletal pathology in CP. Musculoskeletal impairments affect many aspects of the child’s physical functioning, limiting their levels of physical activity and participation (Graham et al. 2016; Rosenbaum et al. 2007b). Children with CP often undergo interventions designed to modify the natural history of musculoskeletal pathologies and improve their gross motor and gait function, including botulinum toxin A injections, physiotherapy, and orthopedic surgery (Graham et al. 2016). It is extremely important to be able to accurately assess the outcomes of these interventions. Outcome measures must consider a child’s level of function across multiple domains (Oeffinger et al. 2009). The World Health Organization’s International Classification of Functioning, Disability and Health (ICF) provides a useful framework for the measurement of outcomes following interventions in children with CP. The ICF considers health conditions with regard to three domains: body structure and function and activities and participation. These domains are influenced by environmental and personal factors (WHO 2001). Various tools exist to measure outcomes relevant to children with CP in the ICF domains, and new measurement tools are being developed. These are shown in Fig. 4. When considering measurement of outcomes in children with CP, it is important to consider all of the components of the ICF. It is also important to choose the correct measurement tool to use at any given time or after an intervention. Choice of measurement tool should be based on the psychometric properties of the tool, the
1062
E. Rutz and P. Thomason
Fig. 4 Measurement tools used for gross motor function in children with CP per ICF domains
aspect of the ICF being measured, as well as the age and GMFCS level of the child. A detailed discussion of measurement tools is beyond the scope of this chapter. Some of the tools that we use commonly are discussed below.
The Diagnostic Matrix Assessment tools can be considered in the context of a diagnostic matrix (Davids et al. 2003). Davids and colleagues described a diagnostic matrix consisting of
Natural History of Cerebral Palsy and Outcome Assessment
1063
clinical history, physical examination, diagnostic imaging, instrumented gait analysis (IGA), and examination under anesthesia as five components of a diagnostic matrix useful in clinical decision making, in relation to gait correction surgery in children with CP. The diagnostic matrix consists of measurements across multiple domains of the ICF and includes many aspects of gross motor function. Incorporating these measurements builds up a comprehensive picture of the individuals’ function which assists with diagnosis and identification of impairments and assist treatment planning especially in relation to major interventions such as gait improvement surgery. Major interventions, such as selective dorsal rhizotomy and multilevel orthopedic surgery, are designed to improve gait and functioning in children with CP. Ideally these major interventions are most appropriately conducted following the most comprehensive and objective assessment possible. Over time we have added additional measurement tools to the diagnostic matrix used in the gait laboratory. Figure. 5 shows the diagnostic matrix and assessment tools included in the matrix.
Body Structure and Function Instrumented Gait Analysis The role of instrumented gait analysis (IGA) is crucial to evaluating gait dysfunction in children with CP, especially in relation to planning and assessing the outcome of major interventions such as selective dorsal rhizotomy (SDR) and single-event multilevel surgery (SEMLS), and is central to the diagnostic matrix. Three-dimensional kinematics, kinetics, and dynamic electromyography provides a comprehensive description of joint movements, moments, and powers and muscle timing which is essential to the management of gait disorders in children with CP (Baker 2013). The objectivity and relative freedom from bias of IGA are factors of major importance in establishing objective outcomes. The use of IGA and a composite measure of gait such as the Gillette Gait Index (Schutte et al. 2000), the Gait Deviation Index (McMulkin and MacWilliams 2015), or the gait profile score (GPS) (Baker et al. 2009) are useful to describe outcomes, of prospective cohort studies as well as randomized trials of multilevel surgery in children with CP. The movement analysis profile (MAP) and overall gait profile score (GPS) have been developed to summarize kinematic data. The root mean square (RMS) difference between nine clinically relevant kinematic variables for a particular child and the average values of that variable from typically developing children are calculated. This represents a clinical meaningful measure of gait variables as it measures difference in degrees and is of value both clinically and in the research setting to evaluate change following surgery (Baker et al. 2009). These measures have revolutionized gait outcome assessment. We are now able to document change in gait, and we can compare kinematic data collected from different centers, allowing the possibility of large cohort collaborative studies.
1064
E. Rutz and P. Thomason
Fig. 5 Assessment Matrix used for assessment of motor function in CP
Video Gait Analysis Video gait analysis (VGA) is a central part of the diagnostic matrix. A visual record of a child’s gait and functioning on digital video can be of much greater value than observational gait analysis and a written report. Digital video is objective and can be shared by multiple observers over a long period. It can be replayed in slow motion and can be reviewed repeatedly. VGA can be used when IGA is either not appropriate or not available. We found VGA particularly useful for the objective documentation of younger children commencing BoNT-A therapy. At this stage, they are too small, too young, and uncooperative for IGA. It is also useful for the selection and monitoring of the use of ankle-foot orthoses (AFOs) and for the monitoring of children after major intervention such as selective dorsal rhizotomy (SDR) or multilevel surgery. In an effort to quantify and objectify the outcome of observational gait analysis, several gait scores have been developed. These include the Physician Rating Scale, the Observational Gait Scale, and the Edinburgh Visual Gait Score (Koman et al. 1993; Mackey et al. 2003; Read et al. 2003; Wren et al. 2005). We do not consider VGA an adequate substitute for IGA when decisions regarding major intervention such as selective dorsal rhizotomy or multilevel surgery have
Natural History of Cerebral Palsy and Outcome Assessment
1065
to be made. Nor is VGA adequate for outcome measurement in clinical trials of gait correction surgery.
Clinical History and Standardized Physical Examination The clinical history is obtained by a careful review of all current and previous medical records complemented by an up to date interview of the child in the context of his family or care providers. Associated medical comorbidities and the response to previous interventions are also crucial to the planning of interventions. Goals and expectations can be established using specific tools such as goal attainment scales, Canadian Occupational Performance Measure and the newly developed Gait Outcomes Assessment List. A standardized routine physical examination is an essential part of the diagnostic matrix. The clinical protocol used in The Hugh Williamson Gait Analysis Laboratory has been published elsewhere (Thomason et al. 2013) and will not be described in detail here. Sagittal Gait Patterns We believe that classification of sagittal gait patterns initially from VGA and then from the information from IGA (sagittal gait kinematics) to be very important. The classification of sagittal gait patterns in spastic hemiplegia (Winters et al. 1987) is a valid and reliable tool which helps in framing logical management strategies. The sagittal gait classification described by Rodda and Graham can be useful in planning intervention in spastic diplegia (Rodda et al. 2004). Both gait classifications suggest common patterns of musculoskeletal deformity and may assist in identifying deformities that require intervention. Standardized Radiology Medical imaging is important in the diagnostic matrix as high proportions of children with CP have skeletal deformities including torsional deformities of long bones and instability of the hip and foot specifically the subtalar and midtarsal joints. Radiology of the hips, including plain radiographs, supplemented by CT measurements of femoral torsion and tibial torsion and more recently, biplanar radiography using EOS (Escott et al. 2013), can be very useful as additional information in the planning of multilevel surgery. Weight-bearing radiographs can be analyzed by the measurement of a series of key radiological indices which can help identify segmental malalignments in the hindfoot, midfoot, and forefoot in a systematic manner (Davids et al. 2005). This contributes greatly to the analysis of segmental foot deformity and the planning of intervention. Instrumented gait analysis most typically interprets the foot as a rigid segment and does not provide detailed information on segmental malalignments within the foot. Until better foot models are in routine use, standardized weight-bearing radiographs remain the cornerstone of analysis of deformities within the foot. This information can be augmented by dynamic pedobarography which is in use in a number of gait laboratories.
1066
E. Rutz and P. Thomason
Longitudinal Assessments with Radiology: Hip Surveillance Centers in Europe and Australia have developed formal “hip surveillance programs” (Dobson et al. 2002; Hagglund et al. 2005; Kentish et al. 2011). Children with a confirmed diagnosis of CP are offered regular clinical and radiographic examination of their hips and access to both preventive and reconstructive surgery. In both Victoria, Australia, and Southern Sweden, the prevalence of late dislocation has decreased, and the need for salvage surgery has been reduced (Dobson et al. 2002; Hagglund et al. 2005). The most useful radiographic index for measuring hip displacement in children with CP is the migration percentage of Reimers 1980. This measures the percentage of the femoral head that lies outside the acetabulum. Migration percentage can be reliably measured from anterior-posterior hip radiographs, taken in supine with good positioning and a standardized technique (Parrott et al. 2002). It is the key index for making decisions about surgical management and to monitor hip displacement both before and after operative intervention. The incidence of hip displacement (migration percentage >30%) is 35% of children with CP in population-based studies and is directly related to the child’s GMFCS level (Soo et al. 2006; Hagglund et al. 2007; Connelly et al. 2009). Early hip displacement is silent, and formal screening by radiographs of the hips is advised. The frequency of such radiographs should be directly related to the risk of hip displacement, which is in turn related to the child’s GMFCS level. Evidence-based recommendations on hip surveillance in children with CP have been published (Wynter et al. 2011).
Activity Functional Mobility Scale The FMS is a six level ordinal scale that rates the mobility of children with CP over three distances according to their need for assistive devices (Graham et al. 2004). The three distances of 5, 50, and 500 Meters represent mobility in the home, school, and wider community settings, respectively. The scale is clinician administered through parent or child report and should reflect performance rather than capability, i.e., what the child actually does do rather than what they can do. For each of the three distances, a rating of 1–6 is assigned. The FMS was designed as an outcome measure and is sensitive to change (Thomason et al. 2011). Following optimum biomechanical realignment and correction of spastic contractures, children can often progress to lesser levels of support than used preoperatively (Thomason et al. 2011). These important changes can be monitored and reported using the FMS (Graham et al. 2004; Harvey et al. 2007).
Functional Assessment Questionnaire The Functional Assessment Questionnaire (FAQ) is a 10-level, parent-reported walking scale, which describes a range of walking abilities across the entire
Natural History of Cerebral Palsy and Outcome Assessment
1067
spectrum of CP, from nonambulatory to independent ambulation at a high level (Novacheck et al. 2000). In addition to the 10-level walking scale, there is an additional list of 22 items describing a variety of higher-level functional activities requiring varying degrees of walking ability, balance, strength, and coordination. The FAQ is a valid and reliable scale and has been shown to be sensitive to change (Gorton et al. 2011). It is a simple scale, which can be quickly completed by parents or caregivers and provides an excellent longitudinal view of the child’s gross motor and walking abilities. It is a good measure of parental perspective however and covers a wide variety of activities of daily living. The FMS and FAQ are complementary scales and are used to assess outcomes in children with CP after intervention.
The Gross Motor Function Measure The Gross Motor Function Measure (GMFM) (Russell and Rosenbaum 1989) is the gold standard for the measurement of gross motor function in children with CP and has been shown to be valid, reliable, and responsive to change (Russell and Rosenbaum 1989; Nordmark et al. 1997, 2000; Bjornson et al. 1998a, b; Russell et al. 2000; Russell and Leung 2003; Russell and Gorter 2005). There are two versions of the GMFM. The original version consisted of 88 items, which were grouped into five dimensions of gross motor function: lying and rolling, sitting, crawling and kneeling, standing and walking, and running and jumping (Russell and Rosenbaum 1989). Following Rasch analysis, a revised version, which consists of 66 items, was developed, is interval scaled, and features item maps. It is quicker to administer. However, it is limited for children who are very young or severely involved (Russell and Leung 2003). The use of GMFM requires an experienced physiotherapist and will take between 45 and 60 min to administer. It is an essential tool in clinical outcome studies to assess change in gross motor function. It is also a useful clinical measure to assess function in CP and guide the management and treatment planning. The Pediatric Evaluation of Disability Inventory (PEDI) (Haley 1997) may be used to assess motor function. The PEDI may be a more appropriate tool for use with children in GMFCS levels IV and V. A detailed examination of the child’s level of activity and participation, using measures such as the Canadian Occupational Performance Measure, Children’s Assessment of Participation and Enjoyment, or the Activity Scale for Kids (Law et al. 1990; Young et al. 2000; King et al. 2004), may be useful. Self-Reported Questionnaires and Health-Related Quality of Life In recent years, the assessment of quality of life has become a major goal in health management including children with CP. A number of generic and specific instruments have been developed which address aspects of health, functioning, and quality of life. The Child Health Questionnaire (CHQ) (Waters et al. 2000) is a widely used tool and is not disease specific. It has the advantage that the scores of children with CP can be compared to children with typical health or with other disease conditions. The Pediatric Orthopaedic Data Collection Instrument (PODCI) (Daltroy et al. 1998)
1068
E. Rutz and P. Thomason
has a more musculoskeletal focus and contains several domains directly relevant to children with CP and gait problems. Although some information exists on the use of both of these instruments in children with CP, the responsiveness to change and the value of using these as outcome instruments are not yet fully established. It is also the case that neither can be considered to be a true quality of life measure. Other questionnaires include the PedsQL and the CPQoL-Child. The CPQoLChild is a specific quality of life measure developed for children with CP; however, its responsiveness to physical interventions such as gait improvement surgery is not yet known (Waters et al. 2006). In order to judge the effectiveness of any intervention in children with CP, it is important to understand the priorities and expectations of the child and parent (Oeffinger et al. 2009; Novak et al. 2012). The Gait Outcomes Assessment List (GOAL) is a new outcome measure to evaluate gait priorities and functional mobility for ambulant children with CP (Narayanan et al. 2015). There are two versions of the questionnaire, parent and child. There are 48 items in both versions grouped into seven domains. Recent research has establishes the validity of the GOAL in measuring the gross motor and gait function of ambulant children with CP. Evidence was found for the discriminative validity of the GOAL, and correlations were demonstrated with standard measures of gross motor function and gait. The GOAL provides meaningful information about a child’s function across multiple dimensions, accounts for the environmental and personal factors that may contribute to function, and measures the priorities and expectations of children and their parents. The GOAL will allow clinicians to better understand the motor abilities, priorities, and expectations of ambulant children with CP and enable better decision making about appropriate interventions. The GOAL will be an invaluable addition to the measurement tools available for gross motor function in CP.
Summary In conclusion, natural history of gait in children with CP is one of deterioration. There is level II evidence that single-event multilevel surgery (SEMLS) improves the gait of children with spastic diplegic CP 12 months after surgery (Thomason et al. 2011). SEMLS (Rutz et al. 2013) results in clinically and statistically significant improvements in gait and function, in children with bilateral spastic CP, which were maintained at 5 years after surgery (Rutz et al. 2013; Thomason et al. 2013), and GMFCS stability was confirmed in the majority of children (Rutz et al. 2012).
Cross-References ▶ Assessing Clubfoot and Cerebral Palsy by Pedobarography ▶ Clinical Gait Assessment by Video Observation and 2D Techniques ▶ Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy
Natural History of Cerebral Palsy and Outcome Assessment
1069
▶ EMG Activity in Gait: The Influence of Motor Disorders ▶ Foot and Ankle Motion in Cerebral Palsy ▶ Functional Effects of Foot Orthoses ▶ Gait Scores: Interpretations and Limitations ▶ Interpreting Ground Reaction Forces in Gait ▶ Interpreting Joint Moments and Powers in Gait ▶ Kinematic Foot Models for Instrumented Gait Analysis ▶ Optimal Control Modeling of Human Movement ▶ Skeletal Muscle Structure in Spastic Cerebral Palsy ▶ Strength Related Stance Phase Problems in Cerebral Palsy ▶ Surface Electromyography to Study Muscle Coordination ▶ Swing Phase Problems in Cerebral Palsy ▶ The Conventional Gait Model - Success and Limitations ▶ Trunk and Spine Models for Instrumented Gait Analysis ▶ Variations of Marker Sets and Models for Standard Gait Analysis
References Baker R (2013) Measuring walking: a handbook of gait analysis. Mac Keith Press, London Baker R, McGinley JL, Schwartz MH et al (2009) The gait profile score and movement analysis profile. Gait Posture 30:265–269 Beckung E, Carlsson G, Carlsdotter S, Uvebrant P (2007) The natural history of gross motor development in children with cerebral palsy aged 1 to 15 years. Dev Med Child Neurol 49(10):751–756 Bell KJ, Ounpuu S, DeLuca PA, Romness MJ (2002) Natural progression of gait in children with cerebral palsy. J Pediatr Orthop 22(5):677–682 Bjornson KF, Graubert CS, Buford VL, McLaughlin J (1998a) Validity of the gross motor function measure. Pediatr Phys Ther 10:43–47 Bjornson KF, Graubert CS, McLaughlin J, Kerfeld CI, Clark EM (1998b) Test retest reliability of the gross motor function measure in children with cerebral palsy. Phys Occup Ther Pediatr 18:51–61 Connelly A, Flett P, Graham HK, Oates J (2009) Hip surveillance in Tasmanian children with cerebral palsy. J Paediatr Child Health 45(7–8):437–443 Daltroy LH, Liang MH, Fossel AH, Goldberg M (1998) The POSNA pediatric musculoskeletal functional health questionnaire: report on reliability, validity and sensitivity to change. The pediatric outcomes instrument development group. J Pediatr Orthop 18:561–571 Davids JR, Ounpuu S, DeLuca PA, Davis RB (2003) Optimization of walking ability of children with cerebral palsy. J Bone Joint Surg Am 85:2224–2234 Davids JR, Gilson T, Pugh LI (2005) Quantitative segmental analysis of weight-bearing radiographs of the foot an ankle for children. J Pediatr Orthop 25(6):769–776 Davids JR, Oeffinger DJ, Bagley AM, Sison-Williamson M, Gorton G (2015) Relationship of strength, weight, age, and function in ambulatory children with cerebral palsy. J Pediatr Orthop 35(5):523–529 Dobson F, Boyd RN, Parrott J, Nattrass GR, Graham HK (2002) Hip surveillance in children with cerebral palsy: impact on the surgical management of spastic hip disease. J Bone Joint Surg Br 84:720–726 Eliasson A-C, Krumlinde-Sundholm L, Rosblad B et al (2006) The manual ability classification system (MACS) for children with cerebral palsy: scale development and evidence of validity and reliability. Dev Med Child Neurol 48:549–554
1070
E. Rutz and P. Thomason
Escott BG, Ravi B, Weathermon AC, Acharva J, Gordon CL, Babyn PS, Kelley SP, Narayanan UG (2013) EOS low-dose radiography: a reliable and accurate upright assessment of lower-limb lengths. J Bone Joint Surg Am 95(23):e1831–e1837 Firth GB, Passmore E, Sangeux M, Thomason P, Rodda J, Donath S, Selber P, Graham HK (2013) Multilevel surgery for equinus gait in children with spastic diplegic cerebral palsy: medium-term follow-up with gait analysis. J Bone Joint Surg Am 95(10):931–938 Gorton GE 3rd, Stout JL, Bagley AM, Bevans K, Novacheck TF, Tucker CA (2011) Gillette functional assessment questionnaire 22.-item skill set: factor and rasch analyses. Dev Med Child Neurol 53:250–255 Graham HK (2004) Mechanisms of deformity. In: Scrutton D, Damiano D, Mayston M (eds) Management of the motor disorders of children with cerebral palsy, Clinics in developmental medicine, vol 161, 2nd edn. Mac Keith Press, London, pp 105–129 Graham HK, Harvey A, Rodda J, Nattrass GR, Pirpiris M (2004) The functional mobility scale (FMS). J Pediatr Orthop 24(5):514–520 Graham HK, Rosenbaum P, Paneth N, Dan B, Lin JP, Damiano DL, Becher JG, Gaebler-Spira D, Colver A, Reddihough DS, Crompton KE, Lieber RL (2016) Cerebral palsy. Nat Rev Dis Primers 2:15082 Gu Y, Shelton JE, Ketchum JM, Cifu DX, Palmer D, Sparkman A, Jermer-Gu MK, Mendigorin M (2011) Natural history of scoliosis in nonambulatory spastic tetraplegic cerebral palsy. PM R 3(1): 27–32 Hagglund G, Wagner P (2008) Development of spasticity with age in a total population of children with cerebral palsy. BMC Musculoskelet Disord 9:150 Hagglund G, Andersson S, Duppe H, Lauge-Pedersen H, Nordmark E, Westbom L (2005) Prevention of dislocation of the hip in children with cerebral palsy. The first ten years of a population-based prevention programme. J Bone Joint Surg Br 87:95–101 Hagglund G, Lauge-Pedersen H, Wagner P (2007) Characteristics of children with hip displacement in cerebral palsy. BMC Musculoskelet Disord 8:101–107 Hagglund G, Alriksson-Schmidt A, Lauge-Pedersen H, Rodby-Bousquet E, Wagner P, Westbom L (2014) Prevention of dislocation of the hip in children with cerebral palsy: 20-year results of a population-based prevention programme. Bone Joint J 96-B(11):1546–1552 Haley SM (1997) The pediatric evaluation of disability inventory. J Rehab Outcome Meas 1:61–69 Hanna SE, Bartlett DJ, Rivard LM, Russell DJ (2008) Reference curves for the gross motor function measure: percentiles for clinical description and tracking over time among children with cerebral palsy. Phys Ther 88:596–607 Harvey A, Graham HK, Morris ME, Baker RJ, Wolfe R (2007) The functional mobility scale: ability to detect changed following single event multilevel surgery. Dev Med Child Neurol 49(8):603–607 Heidt C, Hollander K, Wawrzuta J, Molesworth C, Willoughby K, Thomason P, Khot A, Graham HK (2015) The radiological assessment of pelvic obliquity in cerebral palsy and the impact on hip development. Bone Joint J 97-B(10):1435–1440 Hidecker MJ, Paneth N, Rosenbaum PL, Kent RD, Little J, Eulenberg JB, Chester K Jr, Johnson B, Michalsen L, Evatt M, Taylor K (2011) Developing and validating the communication function classification system (CFCS) for individuals with cerebral palsy. Dev Med Child Neurol 53:704–710 Johnson DC, Damiano DL, Abel MF (1997) The evolution of gait in childhood and adolescent cerebral palsy. J Pediatr Orthop 17(3):392–396 Kentish M, Wynter M, Snape N, Boyd R (2011) Five-year outcome of state-wide hip surveillance of children and adolescents with cerebral palsy. J Pediatr Rehabil Med 4:205–217 King G, Law M, King S et al (2004) Children’s assessment of participation and enjoyment (CAPE) and preferences for activities of children (PAC). Harcourt Assessment, San Antonio Knapp DR Jr, Cortes H (2002) Untreated hip dislocation in cerebral palsy. J Pediatr Orthop 22(5): 668–671 Koman LA, Mooney JF, Smith B (1993) Management of cerebral palsy with botulinum-A toxin: preliminary investigation. J Pediatr Orthop 4:489–495
Natural History of Cerebral Palsy and Outcome Assessment
1071
Law M, Baptiste S, McColl M, Opzoomer A, Polatajko H, Pollock N (1990) The Canadian occupational performance measure: an outcome measure for occupational therapy. Can J Occup Ther 57:82–87 Mackey AH, Lobb GL, Walt SE, Stott NS (2003) Reliability and validity of the observational gait scale in children with spastic diplegia. Dev Med Child Neurol 45(1):4–11 McCormick A, Brien M, Plourde J, Wood E, Rosenbaum P, McLean J (2007) Stability of the gross motor function classification system in adults with cerebral palsy. Dev Med Child Neurol 49:265–269 McGinley J, Dobson F, Ganeshalingam R, Shore BJ, Rutz E, Graham HK (2012) Single-event multilevel surgery for children with cerebral palsy: a systematic review. Dev Med Child Neurol 54:117–128 McMulkin ML, MacWilliams BA (2015) Application of the Gillette gait index, gait deviation index and gait profile score to multiple clinical pediatric populations. Gait Posture 41:608–612 Moreau M, Drummond DS, Rogala E, Ashworth A, Porter T (1979) Natural history of the dislocated hip in spastic cerebral palsy. Dev Med Child Neurol 21(6):749–753 Morgan P, McGinley J (2014) Gait function and decline in adults with cerebral palsy: a systematic review. Disabil Rehabil 36(1):1–9 Murnaghan ML, Simpson P, Robin JG, Shore BJ, Selber P, Graham HK (2010) The cerebral palsy hip classification is reliable: an inter- and intra-observer reliability study. J Bone Joint Surg Am 92:436–441 Narayanan U, Moline R, Encisa C, Yeung C, Weir S (2015) Validation of the GOAL questionnaire: an outcome measure for ambulatory children with cerebral palsy. Dev Med Child Neurol. 57(29) https://doi.org/10.1111/dmcn.45_12887 Nordmark E, Hagglund G, Jarnlo GB (1997) Reliability of the gross motor function measure in cerebral palsy. Scand J Rehabil Med 29:25–28 Nordmark E, Jarnlo GB, Hagglund G (2000) Comparison for the gross motor function measure and paediatric evaluation of disability inventory in assessing motor function in children undergoing selective dorsal rhizotomy. Dev Med Child Neurol 42:245–252 Norlin R, Odenrick P (1986) Development of gait in spastic children with cerebral palsy. J Pediatr Orthop 6(6):674–680 Novacheck TF, Stout JL, Tervo R (2000) Reliability and validity of the Gillette functional assessment questionnaire as an outcome measure in children with walking disabilities. J Pediatr Orthop 20:75–81 Novak I, Hines M, Goldsmith S, Barclay R (2012) Clinical prognostic messages from a systematic review on cerebral palsy. Pediatrics 130(5):e1285–312. Oeffinger DJ, Rogers SP, Bagley A, Gorton G, Tylkowski CM (2009) Clinical applications of outcome tools in ambulatory children with cerebral palsy. Phys Med Rehabil Clin N Am 20:549–565 Palisano R, Rosenbaum P, Walter S, Russell D, Wood E, Galuppi B (1997) Development and reliability of a system to classify gross motor function in children with cerebral palsy. Dev Med Child Neurol 39(4):214–223 Palisano RJ, Rosenbaum P, Bartlett D, Livingston MH (2008) Content validity of the expanded and revised gross motor function classification system. Dev Med Child Neurol 50:744–750 Parrott J, Boyd RN, Dobson F et al (2002) Hip displacement in spastic cerebral palsy: repeatability of radiologic measurement. J Pediatr Orthop 22:660–667 Pritchett JW (1983) The untreated unstable hip in severe cerebral palsy. Clin Orthop Relat Res 173:169–172 Pruszczynski B, Sees J, Miller F (2015) Risk factors for hip displacement in children with cerebral palsy: systematic review. J Pediatr Orthop 36(8):829–833 Read HS, Hazlewood ME, Hillman SJ, Prescott RJ, Robb JE (2003) Edinburgh visual gait score for use in cerebral palsy. J Pediatr Orthop 23:296–301 Reimers J (1980) The stability of the hip in children: a radiological study of the results of muscle surgery in cerebral palsy. Acta Orthop Scand Suppl 184:1–100
1072
E. Rutz and P. Thomason
Robin J, Graham HK, Selber P, Dobson F, Smith K, Baker R (2008) Proximal femoral geometry in cerebral palsy. A population-based cross-sectional study. J Bone Joint Surg Br 90:1372–1379 Rodda JM, Graham HK et al (2004) Sagittal gait patterns in spastic diplegia. J Bone Joint Surg Br 86(2):251–258 Rodda JM, Graham HK, Nattrass GR, Galea MP, Baker R, Wolfe R (2006) Correction of severe crouch gait in patients with spastic diplegia with use of multilevel orthopaedic surgery. J Bone Joint Surg Am 88(12):2653–2664 Rose GE, Lightbody KA, Ferguson RG, Walsh JC, Robb JE (2010) Natural history of flexed knee gait in diplegic cerebral palsy evaluated by gait analysis in children who have not had surgery. Gait Posture 31(3):351–354 Rosenbaum PL, Walter SD, Hanna SE, Palisano RJ, Russell DJ, Raina P, Wood E, Bartlett DJ, Galuppi BE (2002) Prognosis for gross motor function in cerebral palsy: creation of motor development curves. JAMA 288:1357–1363 Rosenbaum P, Paneth N, Leviton A et al (2007a) A report: the definition and classification of cerebral palsy April 2006. Dev Med Child Neurol Suppl 109:8–14 Rosenbaum PL, Livingston MH, Palisano RJ, Galuppi BE, Russell DJ (2007b) Quality of life and health-related quality of life of adolescents with cerebral palsy. Dev Med Child Neurol 49:516–521 Russell DJ, Gorter JW (2005) Assessing functional differences in gross motor skills in children with cerebral palsy who use an ambulatory aid or orthoses: can the GMFM-88 help? Dev Med Child Neurol 47:462–467 Russell DJ, Leung KM (2003) Accessibility and perceived clinical utility of the GMFM-66: evaluating therapists’ judgements of a computer-based scoring program. Phys Occup Ther Pediatr 23(2):45–58 Russell DJ, Rosenbaum PL (1989) The gross motor function measure: a means to evaluate the effects of physical therapy. Dev Med Child Neurol 31(3):341–352 Russell DJ, Avery LM, Rosenbaum PL, Raina PS, Walter SD, Palisano RJ (2000) Improved scaling of the gross motor function measure for children with cerebral palsy: evidence of reliability and validity. Phys Ther 80:873–885 Rutz E (2012) Are hips stable in children with cerebral palsy? Dev Med Child Neurol 54(10):878 Rutz E, Tirosh O, Thomason P, Barg A, Graham HK (2012) Stability of the gross motor function classification system after single-event multilevel surgery in children with cerebral palsy. Dev Med Child Neurol 54(12):1109–1113 Rutz E, Donath S, Tirosh O, Graham HK, Baker R (2013) Explaining the variability improvements in gait quality as a result of single event multi-level surgery in cerebral palsy. Gait Posture 38(3): 455–460 Rutz E, Vavken P, Camathias C, Haase C, Juenemann S, Brunner R (2015) Long-term results and outcome predictors in one-stage hip reconstruction in children with cerebral palsy. J Bone Joint Surg Am 97(6):500–506 Saito N, Ebara S, Ohotsuka K, Kumeta H, Takaoka K (1998) Natural history of scoliosis in spastic cerebral palsy. Lancet 351(9117):1687–1692 Samilson RL, Tsou P, Aamoth G, Green WM (1972) Dislocation and subluxation of the hip in cerebral palsy. Pathogenesis, natural history and management. J Bone Joint Surg Am 54(4): 863–873 Schutte LM, Narayanan U, Stout JL, Selber P, Gage JR, Schwartz MH (2000) An index for quantifying deviations from normal gait. Gait Posture 11:25–31 Soo B, Howard JJ, Boyd RN, Reid SM, Lanigan A, Wolfe R, Reddihough D, Graham HK (2006) Hip displacement in cerebral palsy. J Bone Joint Surg Am 88(1):121–129 Terjesen T (2012) The natural history of hip development in cerebral palsy. Dev Med Child Neurol 54(10):951–957 Thomason P, Graham HK (2013) Rehabilitation of cerebral palsy. In: Iansek R, Morris M (eds) Rehabilitation in movement disorders. Cambridge University Press, Cambridge, UK
Natural History of Cerebral Palsy and Outcome Assessment
1073
Thomason P, Baker R, Dodd K, Taylor N, Selber P, Wolfe R, Graham HK (2011) Single-event multilevel surgery in children with spastic diplegia: a pilot randomized controlled trial. J Bone Joint Surg Am 93(5):451–460 Thomason P, Selber P, Graham HK (2013) Single event multilevel surgery in children with bilateral spastic cerebral palsy: a 5 year prospective cohort study. Gait Posture 37(1):23–28 Waters E, Salmon L, Wake M, Hesketh K, Wright M (2000) The child health questionnaire in Australia: reliability, validity and population means. Aust N Z J Public Health 24:207–210 Waters E, Davis E, Boyd R, Reddihough D, Mackinnon A, Graham HK, Lo SK, Wolfe R, Stevenson R, Bjornson K, Blair E, Ravens-Sieberer U (2006) Cerebral palsy quality of life questionnaire for children (CP QOL-child) manual. Deakin University, Melbourne Wawrzuta J, Willoughby KL, Molesworth C, Ang SG, Shore BJ, Thomason P, Graham HK (2016) Hip health at skeletal maturity: a population-based study of young adults with cerebral palsy. Dev Med Child Neurol 58(12):1273–1280 Winters TF, Gage JR, Hicks R (1987) Gait patterns in spastic hemiplegia in children and young adults. J Bone Joint Surg Am 69:437–441 World Health Organization (2001) International classification of functioning, disability and health. World Health Organization, Geneva, pp 121–160; cited 2016. http://www.who.int/classifica tions/icf/en/ Wren TAL, Rethlefsen SA, Healy BS, Do KP, Dennis SW, Kay RM (2005) Reliability and validity of visual assessments of gait using a modified physician rating scale for crouch and foot contact. J Pediatr Orthop 25:646–650 Wynter M, Gibson N, Kentish M, Love S, Thomason P, Graham HK (2011) The consensus statement on hip surveillance for children with cerebral palsy; Australian standards of care. J Pediatr Rehab Med 4:183–195 Young N, Williams JI, Yoshida KK, Wright JG (2000) Measurement properties of the activities scale for kids. J Clin Epidemiol 53:125–137
Skeletal Muscle Structure in Spastic Cerebral Palsy Adam Shortland
Abstract
The structure of skeletal muscle in cerebral palsy (CP) is altered at the molecular level, at the cellular level, and at the level of the tissue. These abnormalities in structure have implications for active and passive muscle performance and for the functional capacity of the individual, particularly in the long term. Appreciating the deficits of muscle structure may well encourage clinicians to focus on muscle growth when managing this group and lead researchers to novel therapeutics targeted at normalizing muscle structure. Keywords
Skeletal muscle • Growth • Functional reserve • Cerebral palsy
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions and Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Implications for Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1075 1077 1086 1086 1087 1087
Introduction Cerebral palsy (CP) is a lifelong condition arising from a brain lesion affecting motor development in early life. The brain injury is considered unchanging but subsequent central nervous system (CNS) and musculoskeletal development is affected. A. Shortland (*) One Small Step Gait Laboratory, Evelina Children’s Hospital, Guy’s and St Thomas’ NHS Foundation Trust, London, UK e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_51
1075
1076
A. Shortland
Muscles are the only tissues in the human body which generate forces and movements. Adequate muscle size and quality in the lower limbs are required to carry out the routine activities of daily life such as walking, getting up from a chair, and climbing the stairs. Of particular concern here is the alteration of the development of muscle which may limit the ability of the person with CP to carry out the functions of daily living that require muscular strength and the coordination of lower limb movement. In this chapter, we will discuss the prerequisites of good long-term muscle function and how these are compromised in the individual with spastic CP. We will consider the failure of muscle development in children with spastic CP as potential significant factor in the decline of these individuals as they pass into adulthood. Muscle is a complex tissue. Each muscle is an organized composite of individual muscle cells (or fibers) bound together within an extracellular matrix formed from different forms of collagen and proteoglycans. Within each muscle fiber are structures which enable its activation by the repeated release and sequestration of calcium into the cytoplasm (the sarcolemma or cell membrane, the t-tubule system, and the sarcoplasmic reticulum). Muscle cells have mitochondria which are responsible for much of the production of adenosine triphosphate (ATP), the currency of energy in the cell. However, the feature of muscle fibers that differentiate them from other cells are the long chains of force-producing units known as sarcomeres. This repeating subdivision of muscle fibers is the only unit in skeletal muscle able to actively develop significant movement and force. Sarcomeres contain parallel interacting macromolecules of myosin and actin. Molecular extensions of the myosin macromolecule, consisting of a long neck and a globular head, form bonds (crossbridges) with regularly spaced sites on the actin long-chain molecule. Relative movement of myosin and actin (and contraction of the sarcomere) occurs on the interaction of available ATP with the myosin neck causing a transformation of chemical into mechanical energy and change in conformation of the myosin neck. Sequestration of calcium and the presence of adenosine diphosphate (ADP) and phosphorus into the sarcoplasm allows the myosin and actin to associate and the myosin head and neck to return to its resting state. Under unloaded conditions, a single cycle of excitation, activation, and contraction should lead to a shortening of the sarcomere of 430 angstroms (the distance between neighboring active sites on the actin molecules in the longitudinal direction of the muscle fiber). A detailed description of the mechanism of contraction of the sarcomere is not warranted in this chapter but the development of active force by an individual sarcomere is proportionate to the number of crossbridges that can be formed which is dependent on the overlap between the actin and myosin long-chain molecules and the relative speed of movement of these molecules. This dependency of force on length and velocity is characteristic of the sarcomere and, by consequence, of the whole muscle. The synchronized contraction of many sarcomeres contributes to the shortening of individual muscle fibers and of the muscle as a whole. Via tendons linking muscles and bones, muscular contractions lead to joint rotation and joint torques.
Skeletal Muscle Structure in Spastic Cerebral Palsy
1077
State of the Art Skeletal muscle has many functions including the regulation of blood sugars and body temperature. It also has a role as a store of energy being able to be broken down in times of famine. But, the key distinguishing feature of muscle is the development of force and movement. To generate the appropriate levels of force over the lifespan, muscles must have the following qualities (or prerequisites). • Functional flexibility: Ability to produce appropriate forces and contractions speeds for a variety of physical tasks. Collectively, muscles must support the body while also being able to rapidly change the configuration of the skeleton. • Structural integrity: Muscles must be able to resist the large forces to which they are exposed to without frequent failure. • Adaptability secondary to changing demands due to exercise and disuse. • Capacity to self-repair or grow rapidly during childhood. In the following sections, we will discuss these ideal muscle properties and see how they may be altered in individuals with spastic CP. The first prerequisite of muscle function is that muscles must be capable of generating forces and velocities for a variety of physical tasks. For successful movement, the nervous system must present the skeletal musculature with a series of activation sequences and sense the resultant joint rotations and joint forces for modifying or correcting the movement. Further, the activated musculature should be capable of developing the forces and velocities necessary to execute the “planned” movements. Both intracellular properties and the gross morphological features of muscle contribute to the development of muscular force and speed. Muscles rely on energy sources for contraction. The speed at which energy sources can be broken down to transform ADP (adenosine diphospate) into ATP and for that ATP to be delivered to the site of the crossbridges and utilized for the transformation of chemical to mechanical energy determines the velocity of contraction and to some extent the specific tension within a muscle (the active force a muscle can produce per unit area). There is certainly a trade-off between muscle speed and endurance. In slow muscle (muscle containing predominantly slow phenotypes of myosin and large numbers of mitochondria), where the conversion of ADP to ATP for contraction largely takes place within the mitochondria and the delivery of ATP to crossbridges occurs at a lower rate, muscular forces develop slowly but may be maintained for long periods. In contrast, in faster muscle, a significant proportion of the reformation of ATP available for muscular contraction takes place in the cytoplasm. Here, large quantities of ATP are made available but only for short periods (maybe 10–30 s of contractions). Furthermore, there are variants (isoforms) of the myosin molecule in faster muscle which allows a more rapid crossbridge cycling and higher velocity contractions. Largely, each skeletal muscle in the lower limb consists of a mixture of slow and fast fiber type (with the honorable exception of the soleus muscle which has a predominance of slow muscle
1078
A. Shortland
A
C
FORCE B
B D
A
C 1.5 D
2.5
4.0 LENGTH (μm)
Fig. 1 The active force-length relationship of the sarcomere. The active force developed depends on the number of crossbridges that can be formed between actin (red thin strand) and the myosin (the thicker blue strand)
fibers). However, they are highly adaptive tissues in which fiber-type transformation can occur in response to changes in levels of activation and tension. For a more detailed description of the crossbridge cycle, please refer to a muscle physiology text. Although much of the attention in muscle studies is focused on the metabolic and molecular properties, gross morphology and architecture has a much stronger influence on muscle speed and strength than fiber type because they determine the serial and parallel arrangement of sarcomeres within a muscle (Lieber and Fridén 2000). To understand the influence of sarcomere arrangement on muscle dynamics, one must have an appreciation of the force generating capacity of the individual sarcomere. Fig. 1 depicts the active force-length relationship of an individual sarcomere. At longer lengths (~4.0 μ), there is little overlap of the actin and myosin molecules, fewer crossbridges are formed, and little force is produced. Similarly, at shorter lengths, the opportunity from crossbridge formation is limited and little active force is developed. The range over which a sarcomere may produce active forces is around 2.4 μ. At maximal overlap of the actin and myosin macromolecules, the greatest number of crossbridges are formed and the force developed is maximal. In normal muscle, we would expect the resting length of the sarcomere (the length adopted in the muscle’s passive state with no external forces applied) to be close to the position of optimal force (at about 2.4 μ). This would allow muscles to produce their maximum forces quickly from rest. The potential significance of an altered sarcomere resting length will become clear when we discuss skeletal muscle in spastic CP. An individual sarcomere develops little force and acts over a very limited range. However, when linked together, they have the capacity to produce the magnitude of forces and speeds that we require to maintain our posture against gravity and to move our joints through significant ranges of motion. Imagine two small (2 dimensional) muscles, each created from just 12 sarcomeres (Fig. 2).
Skeletal Muscle Structure in Spastic Cerebral Palsy
1079
Muscle A
FORCE
Muscle A Muscle B
Muscle B LENGTH
Fig. 2 The effect of sarcomere arrangement on muscle force development
Each sarcomere can produce the same peak forces, has the same active range, and has the same maximal velocity of contraction. In muscle A, the sarcomeres are arranged with 4 in parallel and 3 in series while in muscle B there are 3 sarcomeres in parallel and 4 in series. Even though these two hypothetical muscles have the same number of sarcomeres, they behave very differently. Muscle A can produce 4/3 times the peak force of muscle B, while muscle B has 4/3 the active range and 4/3 the speed of muscle A. In brief, the number of sarcomeres acting in parallel within a muscle determines its peak active force while the number of sarcomeres acting in series determines its active range and speed. If we scale up these tiny muscles to something more approaching the sizes of those of the human body, we can appreciate that the peak active force developed by a muscle is proportionate to the area of the muscle offered in the direction of its external tendon and the range and velocity of the muscle is proportionate to its fiber type. Of course, these relationships assume that the area of the muscle is largely composed of sarcomeric (or myofibrilar) tissue and that the force length relationship of the sarcomere is homogenous throughout the muscle. There are few muscles in the human body that have the linear appearance of muscles A or B. Most muscles have a pennate structure in which muscle fibers make an angle with the tendons with which they are joined. This arrangement is thought to allow the efficient packing of sarcomeres to optimize force production and active range. Fukanaga and his colleagues (Fukunaga et al. 1992) used a simple mathematical expression (Eq. 1) in units of area which is proportionate to the number of sarcomeres acting in parallel in a pennate muscle in the direction of the tendon which
1080
A. Shortland
they termed the physiological cross-sectional area (PCSA). They were able to show that for individual muscle PCSA is a strong linear predictor of peak muscular force. Physiological Cross-Sectional Area PCSA ¼
V cos θ lf
(1)
where V is the muscle volume, θ is the angle of pennation and lf is the length of the muscle fibres. Some interesting studies of the morphology and architecture of the lower limbs in typically developing adult subjects have been conducted. Notably, and most recently, Ward and coworkers (2009) analyzed gross muscle structure from a large number of cadavers. It is clear that function follows form in the human musculature with antigravity muscles (such as the soleus) having large PCSAs and relatively short muscle fibers while the flexors of the limb (such as the ankle dorsiflexors) have longer fiber lengths but smaller PCSAs. A limited number of studies have been conducted of the metabolic phenotype and cross-sectional shape and size of muscle fibers in spastic CP. Ito et al. found fibers to be predominantly slow in the lower limb muscles of ambulant individuals with increased variation in fiber cross-sectional area (Rose et al. 1994; Araki et al. 1996). The largest number of samples reported in a single study were taken by Castle and colleagues (1979). These were taken from multiple muscles in the lower limb and showed a large heterogeneity in muscle fiber type, increased variation in fiber diameter. It should be stated that these studies were poorly controlled (it is difficult to find samples from representative muscles in control subjects) or that the studies were of limited sample sizes, and sometimes, did not fully take into account the clinical presentation of the individuals from which the samples were taken, and that the samples were taken from upper and lower limbs. More recently, in samples of tissue taken from the semitendinosis of children with CP and those from children with ACL injury about to have surgeries suggested a predominance of fibers with a slower phenotype and a reduced cross-sectional area (Zogby et al. 2016). Gantelius and colleagues (2012) found muscles in the forearms of children with unilateral CP to have a greater expression of myosin heavy chain IIx, consistent with a faster fiber phenotype These studies collectively show alterations in muscles at the level of the fiber in CP that are not consistent with simple model of chronic over-activity (where a uniform slow phenotype may emerge) or a model of disuse (where we may expect faster smaller fibers to be present) (Foran et al. 2005). In contrast to the variable results reported for muscle fiber type, a consistent feature of spastic CP is reduction in lower limb muscle size. There is a lack of data for muscles of the upper limb, but the majority of muscles in the lower limb are affected with reductions in muscle volume (normalized to muscle mass) of between 20% and 50% reported (Noble et al. 2014a; Fry et al. 2007; Shortland 2009; Barber et al. 2011; Handsfield et al. 2016). These large reductions in muscle volume are likely to have a large effect on the functional capacity of an individual with
Skeletal Muscle Structure in Spastic Cerebral Palsy
1081
spastic CP. Musculoskeletal simulations suggest that muscle volume deficits of 40%-60% would severely compromise ambulatory function in children with unimpaired gait moderate crouch (Steele et al. 2012). A global reduction in muscle volume would be an interesting enough finding, however, in those studies, where a near complete set of lower limb muscle volumes have been measured (Noble et al. 2014a; Handsfield et al. 2016; Lampe et al. 2006) the distal musculature appear more affected while certain muscles such as the quadriceps appear relatively spared. This is an interesting observation from two perspectives. Firstly, an uneven distribution of muscle deficits contradicts the notion of a disuse model of muscle atrophy while greater distal involvement implies a neurological influence on muscle development (Gough and Shortland 2012). Herskind and colleagues recently demonstrated that gastrocnemius muscle volumes and cross-sectional areas were reduced in children below the age of 18 months implying that muscles of infants with CP are on a different developmental track with reduced rates of growth compared to their typically developing peers (Herskind et al. 2016). Noble et al. (in review) show similar results in a cross-sectional study of adolescents and young adults, reporting a disparity in rate of muscle growth in relation to the accrual of bodyweight. Secondly, the distal musculature is very important for standing and walking in typically developing subjects. Larger deficits in the distal musculature of individuals in CP may particularly compromise these activities. Greater muscle deficits in children with higher gross motor functional classification system (GMFCS) scores (Noble et al. in review) gives further support to the idea that reduction in muscle volume are related to ambulatory mobility. However, studies with greater numbers and a wider range of GMFCS levels are required to fully map out the associations between muscle volume and functional capacity. Further, longitudinal studies are required to elaborate any causal relationships between reduced muscle growth rates and trajectories of functional improvement and decline. There continues to be a dearth of studies of muscle volume in the upper limb. The active range of muscles depend on the fiber length of muscles and on the number of sarcomeres per unit fiber length. It is difficult to measure the length of fibers directly but B-mode ultrasound imaging allows the length of bundles of fibers (or fascicles) to be estimated (Fig. 3). It had been a long-held belief that the cause of muscular deformity (normally, manifesting itself as a limitation in passive joint range) is short muscle fibers. However, the results of imaging studies are rather equivocal (Barber et al. 2011; Malaiya et al. 2007; Khan et al. 2008; Mathewson et al. 2015) with some workers claiming no measureable difference of the length of fascicles between children with spastic CP and control subjects while others have measured small reductions in fascicle length in children with CP. There are few studies reporting measurements from muscles other than the plantarflexors, probably, because of the technical difficulties associated with imaging longer fascicle lengths with standard 2D ultrasound probes. A notable exception is from the work of Moreau et al. (2009) in which the researchers measured fascicle lengths and fascicle angles in the quadriceps of children with CP. Shortness of the muscle fibers, in itself, now seems unlikely to be a significant contributor to measured passive muscle stiffness. Recently, Smith et al. (2011)
1082
A. Shortland
Fig. 3 Composite longitudinal ultrasound image of the tibialis anterior in a typically developing adult subject. As indicated by the red lines, fascicle length and fascicle angle may be measured directly from the ultrasound images
demonstrated that the increased stiffness of hamstring muscles in children with CP having surgery to lengthen these muscles had increased quantities of extracellular matrix (ECM) and that the increased Young’s modulus associated with this extra material could account for increased stiffness of fascicular bundles of tissue. Perhaps, the more extraordinary finding from this paper was that the authors found resting sarcomere lengths that were longer than in control subjects (children having ACL repairs). Certainly, sarcomeres being close to the limits of their extension (at about 4 μ) would contribute to the passive resistance to stretch but the implications of this finding may be more wide-reaching, affecting the active properties of the muscle. If the resting length of the sarcomere is increased, then the overlap between the sarcomeric macromolecules actin and myosin would be reduced with a consequent reduction in the number of crossbridges formed and force produced. In order to reach optimal sarcomere lengths to produce maximal forces, each sarcomere would need to contract by more than a micron (Fig. 4). If the sarcomere contracted by the distance between each active site on the actin molecule per activation cycle, then it would take the sarcomere about 20 activation cycles to go from its resting state to its optimal state. Under those conditions, the muscle would generate its maximum force from rest at more than half a second after it was initially activated. Downing and colleagues (2009) demonstrated that the time to maximal activation was more than 0.5 s in a group of children with CP, consistently twice as slow as their able-bodied peers. The long resting length of sarcomeres may also have implications for the joint position at which optimal joint torques are developed. Matthewson et al. (2015) estimated the number of sarcomeres in series in the soleus muscle. She found resting sarcomere lengths of around 4 μ in children about to undergo surgery in contrast to lengths of 2.17 μ in typically developing children. With similar fascicle lengths in both groups as measured by ultrasound, one can estimate the number of sarcomeres in series (about 10,000). If each sarcomere has to contract by 1.5 μm to operate at its optimal force, then the soleus would need to contract by 1.5 cm from its resting length. If we assume a moment arm of the Achilles tendon at the ankle of 4 cm, then the ankle joint would need to plantarflex by 21 degrees. Could the disparity between the active and passive force length properties of the sarcomere in children with spastic CP explain phenomena such as
Skeletal Muscle Structure in Spastic Cerebral Palsy Fig. 4 A schematic of the resting length of sarcomeres in typically developing subjects and subjects with spastic cerebral palsy. The greater resting length in subjects with CP suggests a reduced overlap between the sarcomeric proteins of actin and myosin implying a greater time period for the muscle to reach its optimal (peak) force from a resting position
1083
3.6m
CP
TD
2.6mm dynamic equinus or crouch gait? At the moment, the extraordinary finding of “overstretched” sarcomeres in the lower limb is limited to a small number of muscles and has been made by a single, albeit, expert group. Certainly, more studies are required if we are to confirm a relationship between overstretched sarcomeres and dynamic shortness in these children. One interesting coda to the story of the overstretched sarcomere is what happens to muscle fascicle length after surgery. Shortland et al. (2004) in the medial gastrocnemius demonstrated reduction in fascicle length after surgery. Such a change could be attributed to sarcomere loss, or it could be that sarcomere length becomes normalized after surgery by a mechanism that relieves passive tension in the muscle. At the moment, we do not know the answer, but the observation poses the question about how normal sarcomere function may be restored. While a number of biopsy studies have found increased ECM in muscles of the upper and lower limbs of children with spastic CP, fewer studies have investigated changes in muscle composition at the level of the whole muscle (Johnson et al. 2009; Noble et al. 2014b). These studies show increased level of subcutaneous, intermuscular and intramuscular fat and raised levels of connective tissue in individuals with bilateral CP. The effect of a raised fraction of nonmyofibrillar tissue would tend to decrease the active specific tension within a muscle and by implication the force developing capacity of that muscle. Our second prerequisite is that muscle should have mechanical integrity. Muscle is a complex composite material. Muscle fibers consist of multiple parallel myofibrils (long chains of sarcomeres). Muscles themselves consist of fibers positioned (largely) in parallel with each other. If these structures produced forces independently or were exposed individually to externally imposed tensile forces, then the specific tensions generated would be highly destructive. Instead, forces within muscles are shared across the muscular components by molecular linkages within and between muscle fibers.
1084
A. Shortland
Fig. 5 (a) The myosin filament is linked to the z-disc of the sarcomere by the large macromolecule titin, providing integrity for sarcomeres in series. (b) Sarcomeres are bound together in parallel at the z-disc by molecules such as α-actinin. (c) Muscle fibers are connected together through a complex of molecules including desmin and dystrophin which join the z-discs with the extracellular matrix
At the level of the sarcomere, there are long chain proteins that prevent dissociation of actin and myosin. Of specific interest is titin. This enormous molecule connects myosin to the z-disc (the stiff network of actin-like molecules that define the ends of each sarcomere. Titin has an elastic subcomponent which allows the molecule to stretch (presumably to allow different degrees of overlap between actin and myosin). This molecule becomes stiff at long sarcomere lengths and therefore is responsible for maintaining serial integrity between sarcomeres along the length of the myofibril (Fig. 5a). But how is mechanical integrity maintained in parallel? The thin filament, actin is bound to the z-disc by the dimer α-actinin. This binding helps to form a 3D stiff structure among neighboring sarcomeres and thus parallel integrity within the muscle fiber is maintained (see Fig. 5b). Mechanical connections between neighboring fibers are maintained though a complex of proteins collectively known as costameres (Fig. 5c) which attach to the z-discs of peripheral sarcomeres within the fiber to sites in the ECM. The combination of these mechanical proteins enables lateral force transmission between neighboring sarcomeres and neighboring muscle fibers allowing the forces generated by the activation of individual muscle fibers to be distributed across the muscle, avoiding local raised stress concentrations. Further, these molecules probably help to regulate sarcomere length across the muscle. There are no reports of significant alterations in these “mechanical” proteins in spastic CP. Indeed, Smith et al. found that titin was not changed and did not contribute to the increased mechanical stiffness of fascicular bundles (bundles of 20 or more fibers) taken from the hamstrings of children with spastic CP just prior to surgery to lengthen these muscles. However, the authors did note that the increased stiffness was associated with increased levels of collagen with the ECM of the bundle. This finding suggests that internal forces are generated between the ECM and the muscle fibers that causes overstretching of sarcomeres. This could be related to differential distances between costameric bindings to the ECM and costameric attachments at the z-discs of the sarcomeres. Our third perquisite for muscle function is adaptability. Muscle is, in fact, a highly adaptable biological material. Cell density is high within muscle and protein turnover is great. This makes muscle a plastic material, responsive to changes in demand.
Skeletal Muscle Structure in Spastic Cerebral Palsy
1085
The rates of sarcomeric protein generation and breakdown are regulated by systemic factors such as myostatin, tumor necrosis factor, and insulin-like growth factor and by additional transcriptional factors that reside within the cell. Until recently, it was not known precisely how mechanical or electrical stimuli were translated into muscle growth. There is an excellent review of muscle adaptability by Braun and Gautel (2011) and the reader is referred here for a deeper understanding. Briefly, transcriptional factors associated with the sarcomere are mobilized by stretching of the sarcomere and these are transported to the nucleoplasm where they become linked with a particular gene associated with the production of an associated protein. Similarly, on activation of a muscle fiber, calcium ions are released which provokes the calcineurin pathway and stimulates production (upregulates) trophic pathways. So, when a muscle is stretched or activated there are molecular signals which stimulate trophic pathways and building of muscle. Equally, there are pathways that depress (downregulate) protein production and cause atrophy. Therefore, muscle hypertrophy and atrophy are directly related to muscle use and disuse. In spastic CP, many of the pathways promoting the building of muscle are depressed and many of the pathways that causes atrophy are activated (Smith et al. 2012). It is likely that the downregulation of these trophic pathways is directly related to the reductions in the rates of muscle growth in spastic CP that are observed macroscopically. Our fourth prerequisite for muscle function concerns the capacity for muscles to repair themselves and to grow rapidly to match growth of the body. Muscles produce large forces when active and necessarily cause large stresses within their tissues. This can lead to cell damage and potential cell death. Muscles have additional cells that during early development were not differentiated to myoblasts (the precursor cells of muscle fibers). These “satellite” cells reside close to muscle fibers (between the cell membrane and the basal lamina). When a muscle cell is damaged, satellite cells proliferate with some differentiating to form myoblasts which fuse with the damage cell and donate their nuclei, while others are dedicated to maintaining the satellite cell population (for review, see (Dayanidhi and Lieber 2014)). The addition of nuclei to an already multinucleated cell allows the cell to recover and grow. In the typically developing mature subject, about 5% of muscle nuclei reside in satellite cells. The capacity of satellite cells to activate or maintain a quiescent state depends upon the signaling milieu within which they exist – the satellite cell niche – thus rates of self-renewal and of functional differentiation may be affected by the properties of tissue in their immediate vicinity including those of the ECM. Conversely, satellite cells modify the ECM and local fibroblast activity. In typically developing infants, the nuclei of satellite cells are thought to make up about 30% of the total muscle nuclei. This high density of stem cells supports an extraordinary rate of muscle growth in early life as these cells donate their nuclei to the developing muscle fibers. Lieber and colleagues found large deficits in the number of satellite cells in the hamstrings of children with spastic cerebral palsy just prior to surgery (Smith et al. 2013). The low number of muscle stem cells may explain the reduced rates of longitudinal and cross-sectional growth in spastic CP31.The reasons for the reduced
1086
A. Shortland
population of SCs are unclear but it may be related to the extensive ECM that is present in these children and also of the ECM’s altered mechanical properties.
Conclusions and Future Directions In summary, the force developing capacity and the speed of muscles in the lower limbs of children with spastic CP are compromised by multiple aberrant structural features including reduced muscle volume (and physiological cross section), reduced serial sarcomere number, altered sarcomere length, and increased nonmyobrillar content. These abnormalities in microscopic and macroscopic properties may be related to a depressed population of satellite cells, those cells that potentiate growth and repair. While researchers and clinicians are getting closer to a fuller understanding of nature of muscular contracture in spastic CP, the origins of these problems are unknown. It is likely, that the developmental trajectory of muscle in affected children is set very early in life when the innervation of skeletal muscle is being organized and refined. Certainly, we understand that that it takes some time for the ratio of myofibrillar tissue to ECM to increase during postnatal growth and that this is related to the sustained activation of the musculature (Eken et al. 2008). In children with spastic CP, there is an injury to the developing brain that compromises the development of the descending neural tracts, particularly of the cortico-spinal tract (Clowry 2007). Among other things, these tracts are responsible for the lowering of the excitation thresholds of alpha motor neurons and the increase in efferent output from the spinal cord to the developing musculature (Lemon 2008). If efferent output from the spinal cord to the musculature in young children with spastic CP is diminished, then it is probable that the development of the myofibrillar fraction is compromised and the ECM is not refined during the first 1 or 2 years of childhood. The maldevelopment of the ECM may well then affect the regulation of sarcomere length and the proliferation and activation of satellite cells. Subsequent muscle growth during childhood would be reduced and the relationship between the active and passive properties of muscles deranged. As yet there is very little in the literature documenting early muscle development in those children at risk of CP18. Certainly, there are significant ethical and logistical problems with sampling muscle tissue from infants. However, there is an opportunity to use simple imaging methods such as echogenicity from B-mode ultrasound to estimate the nonmyofibrillar fraction in muscles during development (Pillen et al. 2003).
Implications for Treatment The altered trajectory of muscle development in spastic CP has important implications for mobility across the lifespan. Shortland (2009) proposed a conceptual framework for understanding the decline of mobility in the affected young adult. In brief, he suggested that due to the reduced rate of accretion of muscle mass, the acquisition of important functional milestones (such as standing walking, getting up
Skeletal Muscle Structure in Spastic Cerebral Palsy
1087
from a chair) would be delayed. Furthermore, he thought that the reduced rates of muscle growth would lead to a reduction in the muscular reserve of an individual (i.e., the muscular forces that the individual could develop above the threshold required to perform the activity). This would result in increased levels of fatigue for the individual performing that activity and reduce the number of years that the individual would be able to perform that activity as they become exposed to the deleterious effects of aging on muscle properties. Clinical management, therefore, should be directed at developing and maintaining a muscular reserve and avoiding interventions that result in long-term atrophy. Studies of resistance training (McNee et al. 2009) and of muscle stimulation suggest that muscle mass can be developed in short-term programs. However, long-term muscle hypertrophy in this group may be limited by the reduced population of satellite cells available to support growth and repair. Increasing the potential for growth and self-repair by cellular and molecular interventions is the holy grail of muscle management in spastic CP, but at the moment, we lack a precise understanding of the mechanism of the development of muscular deformity that might be challenged by early molecular intervention.
Cross-References ▶ 3D Musculoskeletal Kinematics Using Dynamic MRI ▶ Cross-Platform Comparison of Imaging Technologies for Measuring Musculoskeletal Motion ▶ EMG Activity in Gait: The Influence of Motor Disorders ▶ Optimal Control Modeling of Human Movement ▶ Spasticity Effect in Cerebral Palsy Gait ▶ Ultrasound Technology for Examining the Mechanics of the Muscle, Tendon, and Ligament
References Barber L, Hastings-Ison T, Baker R, Barrett R, Lichtwark G (2011) Medial gastrocnemius muscle volume and fascicle length in children aged 2 to 5 years with cerebral palsy. Dev Med Child Neurol 53(6):543–548. https://doi.org/10.1111/j.1469-8749.2011.03913.x Braun T, Gautel M (2011) Transcriptional mechanisms regulating skeletal muscle differentiation, growth and homeostasis. Nat Rev Mol Cell Biol 12(6):349–361. https://doi.org/10.1038/ nrm3118 Castle ME, Reyman TA, Schneider M (1979) Pathology of spastic muscle in cerebral palsy. Clin Orthop Relat Res 142:223–232. http://www.ncbi.nlm.nih.gov/pubmed/159152. Accessed 1 Apr 2012 Clowry GJ (2007) The dependence of spinal cord development on corticospinal input and its significance in understanding and treating spastic cerebral palsy. Neurosci Biobehav Rev 31(8):1114–1124. https://doi.org/10.1016/j.neubiorev.2007.04.007 Dayanidhi S, Lieber RL (2014) Skeletal muscle satellite cells: mediators of muscle growth during development and implications for developmental disorders. Muscle Nerve 50(5):723–732. https://doi.org/10.1002/mus.24441
1088
A. Shortland
Downing AL, Ganley KJ, Fay DR, Abbas JJ (2009) Temporal characteristics of lower extremity moment generation in children with cerebral palsy. Muscle Nerve 39(6):800–809. https://doi. org/10.1002/mus.21231 Eken T, Elder GCB, Lømo T (2008) Development of tonic firing behavior in rat soleus muscle. J Neurophysiol 99(4):1899–1905. https://doi.org/10.1152/jn.00834.2007 Foran JRH, Steinman S, Barash I, Chambers HG, Lieber RL (2005) Structural and mechanical alterations in spastic skeletal muscle. Dev Med Child Neurol 47(10):713–717. https://doi.org/ 10.1017/S0012162205001465 Fry NR, Gough M, McNee AE, Shortland AP (2007) Changes in the volume and length of the medial gastrocnemius after surgical recession in children with spastic diplegic cerebral palsy. J Pediatr Orthop 27(7):769–774. https://doi.org/10.1097/BPO.0b013e3181558943 Fukunaga T, Roy RR, Shellock FG et al (1992) Physiological cross-sectional area of human leg muscles based on magnetic resonance imaging. J Orthop Res 10(6):928–934. https://doi.org/ 10.1002/jor.1100100623 Gantelius S, Hedström Y, Pontén E (2012) Higher expression of myosin heavy chain IIx in wrist flexors in cerebral palsy. Clin Orthop Relat Res. https://doi.org/10.1007/s11999-011-2035-3 Gough M, Shortland AP (2012) Could muscle deformity in children with spastic cerebral palsy be related to an impairment of muscle growth and altered adaptation? Dev Med Child Neurol 54(6):495–499. https://doi.org/10.1111/j.1469-8749.2012.04229.x Handsfield GG, Meyer CH, Abel MF, Blemker SS (2016) Heterogeneity of muscle sizes in the lower limbs of children with cerebral palsy. Muscle Nerve 53(6):933–945. https://doi.org/ 10.1002/mus.24972 Herskind A, Ritterband-Rosenbaum A, Willerslev-Olsen M et al (2016) Muscle growth is reduced in 15-month-old children with cerebral palsy. Dev Med Child Neurol 58(5):485–491 Ito J, Araki A, Tanaka H, Tasaki T, Cho K, Yamazaki R (1996) Muscle histopathology in spastic cerebral palsy. Brain and Development. https://doi.org/10.1016/0387-7604(96)00006-X Johnson DL, Miller F, Subramanian P, Modlesky CM (2009) Adipose tissue infiltration of skeletal muscle in children with cerebral palsy. J Pediatr 154(5):715–720. https://doi.org/10.1016/j. jpeds.2008.10.046 Mohagheghi AA, Khan T, Meadows TH, Giannikas K, Baltzopoulos V, Maganaris CN (2008) In vivo gastrocnemius muscle fascicle length in children with and without diplegic cerebral palsy. Dev Med Child Neurol 50(1):44–50. https://doi.org/10.1111/j.1469-8749.2007.02008.x Lampe R, Grassl S, Mitternacht J, Gerdesmeyer L, Gradinger R (2006) MRT-measurements of muscle volumes of the lower extremities of youths with spastic hemiplegia caused by cerebral palsy. Brain and Development 28(8):500–506. https://doi.org/10.1016/j.braindev.2006.02.009 Lemon RN (2008) Descending pathways in motor control. Annu Rev Neurosci 31:195–218. https:// doi.org/10.1146/annurev.neuro.31.060407.125547 Lieber RL, Fridén J (2000) Functional and clinical significance of skeletal muscle architecture. Muscle Nerve 23:1647–1666. https://doi.org/10.1002/1097-4598(200011)23:113.0.CO;2-M. [pii] Malaiya R, McNee AE, Fry NR, Eve LC, Gough M, Shortland AP (2007) The morphology of the medial gastrocnemius in typically developing children and children with spastic hemiplegic cerebral palsy. J Electromyogr Kinesiol 17(6):657–663. https://doi.org/10.1016/j.jelekin.2007. 02.009 Mathewson MA, Ward SR, Chambers HG, Lieber RL (2015) High resolution muscle measurements provide insights into equinus contractures in patients with cerebral palsy. J Orthop Res. https:// doi.org/10.1002/jor.22728 McNee AE, Gough M, Morrissey MC, Shortland AP (2009) Increases in muscle volume after plantarflexor strength training in children with spastic cerebral palsy. Dev Med Child Neurol 51(6):429–435. https://doi.org/10.1111/j.1469-8749.2008.03230.x Moreau NG, Teefey SA, Damiano DL (2009) In vivo muscle architecture and size of the rectus femoris and vastus lateralis in children and adolescents with cerebral palsy. Dev Med Child Neurol 51(10):800–806
Skeletal Muscle Structure in Spastic Cerebral Palsy
1089
Noble JJ, Chruscikowski E, Fry NR, Lewis AP, Gough M, Shortland AP. Reduced lower limb muscle growth in relation to body mass in a cross-sectional study of ambulant individuals with bilateral cerebral palsy aged 10 to 23. Article in review Noble JJ, Fry NR, Lewis AP, Keevil SF, Gough M, Shortland AP (2014a) Lower limb muscle volumes in bilateral spastic cerebral palsy. Brain and Development 36:294–300. https://doi.org/ 10.1016/j.braindev.2013.05.008 Noble JJ, Charles-Edwards GD, Keevil SF, Lewis AP, Gough M, Shortland AP (2014b) Intramuscular fat in ambulant young adults with bilateral spastic cerebral palsy. BMC Musculoskelet Disord 15:236. https://doi.org/10.1186/1471-2474-15-236 Pillen S, Scholten RR, Zwarts MJ (2003) Verrips a. Quantitative skeletal muscle ultrasonography in children with suspected neuromuscular disease. Muscle Nerve 27(6):699–705. https://doi.org/ 10.1002/mus.10385 Rose J, Haskell WL, Gamble JG, Hamilton RL, Brown DA, Rinsky L (1994) Muscle pathology and clinical measures of disability in children with cerebral palsy. J Orthop Res 12(6):758–768. https://doi.org/10.1002/jor.1100120603 Shortland A (2009) Muscle deficits in cerebral palsy and early loss of mobility: can we learn something from our elders? Dev Med Child Neurol 51(Suppl 4):59–63. https://doi.org/10.1111/ j.1469-8749.2009.03434.x Shortland AP, Fry NR, Eve LC, Gough M (2004) Changes to the muscle architecture of the medial gastrocnemius after surgical intervention in spastic diplegia. Dev Med Child Neurol 46:667–673 Smith LR, Lee KS, Ward SR, Chambers HG, Lieber RL (2011) Hamstring contractures in children with spastic cerebral palsy result from a stiffer extracellular matrix and increased in vivo sarcomere length. J Physiol 589(Pt 10):2625–2639. https://doi.org/10.1113/jphysiol.2010. 203364 Smith LR, Chambers HG, Subramaniam S, Lieber RL (2012) Transcriptional abnormalities of hamstring muscle contractures in children with cerebral palsy. PLoS One 7(8):e40686. https:// doi.org/10.1371/journal.pone.0040686 Smith LR, Chambers HG, Lieber RL (2013) Reduced satellite cell population may lead to contractures in children with cerebral palsy. Dev Med Child Neurol 55:264–270. https://doi. org/10.1111/dmcn.12027 Steele KM, van der Krogt MM, Schwartz MH, Delp SL (2012) How much muscle strength is required to walk in a crouch gait? J Biomech 45(15):2564–2569. https://doi.org/10.1016/j. jbiomech.2012.07.028 Ward SR, Eng CM, Smallwood LH, Lieber RL (2009) Are current measurements of lower extremity muscle architecture accurate? Clin Orthop Relat Res 467:1074–1082. https://doi. org/10.1007/s11999-008-0594-8 Zogby AM, Dayanidhi S, Chambers HG, Schenk S, Lieber RL (2016) Skeletal muscle fiber-type specific succinate dehydrogenase activity in cerebral palsy. Muscle Nerve. https://doi.org/ 10.1002/mus.25379
Part XIV Movement Deviations in Cerebral Palsy
Swing Phase Problems in Cerebral Palsy Ana Presedo
Abstract
The normal human gait cycle is commonly divided in two phases: stance and swing. The objective of stance phase is to provide support, stability, and propulsion and contribute to the advancement of the limb in swing. Ground clearance and appropriate pre-positioning of the foot are prerequisites of normal gait during swing. Normal characteristics of stance and swing phases ensure adequate step length and energy conservation during walk. Swing phase problems in cerebral palsy tend to be related to spasticity, abnormal muscular control, poor opposite limb balance, and lack of power. From a clinical point of view, knee stiffness and ankle insufficient dorsiflexion constitute the two major abnormalities in the sagittal plane, whereas in the frontal and transverse planes, excessive hip adduction and foot deviations can cause problems with limb advancement. Rectus femoris (RF) abnormal activity has been proposed as the main cause of knee stiffness in patients with cerebral palsy. Spasticity and/or contractures of the hip adductors and tibialis posterior muscles can cause abnormal hip and ankle kinematics and can lead to foot deformities if left untreated. Indications for treatment should be based on patient’s age and functional status. Different therapeutic approaches are described and commented in this chapter. A review of the literature based on the major articles in this field is provided. Keywords
Swing phase • Stiff-knee gait • Rectus femoris release • Split anterior tibial tendon transfer
A. Presedo (*) Pediatric Orthopaedics Department, Robert Debré University Hospital, Paris, France e-mail: [email protected]; [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_53
1093
1094
A. Presedo
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Swing Phase Abnormal Kinematic Patterns and Causes of Pathological Function . . . . . . . . . . . Hip and Pelvis Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Knee Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ankle and Foot Pathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Management of Swing Phase Problems in Cerebral Palsy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Excessive Hip Adduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stiff-Knee Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Insufficient Ankle Dorsiflexion and Foot Varus Deviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1094 1095 1096 1096 1098 1101 1103 1103 1103 1105 1106 1107
Introduction Swing phase represents about 40% of the gait cycle. The purposes of swing are to advance the limb, to provide foot clearance, to allow variation in cadence, and to participate to the energy conservation mechanisms. The swing phase is commonly subdivided in three periods: initial swing, mid-swing, and terminal swing. The objective of the first period is to accelerate the limb, and this function is dependent on muscular work occurring during terminal stance and pre-swing, the second period represents a transition period, and the third period’s functional goal is to decelerate the limb to prepare it for initial contact. At normal self-selected walking speed, the limb swings with little muscle intervention; only the tibialis anterior is active throughout swing (Piazza and Delp 1996). In terminal swing, hamstring contraction generates simultaneous hip extension and knee flexion. The action of hip extension contributes to level the pelvis and to maintain trunk alignment, whereas the knee flexion moment decelerates the shank and controls knee extension at initial contact. The main role of the rectus femoris (RF) muscle in normal gait is to act as a hip flexor to propel the limb forward into swing (Perry 1992). Using fire-wire electrodes, two bursts of activity have been described (Annaswamy et al. 1999). The first burst occurs during the loading response phase of gait where the RF acts along with the vastii, and the second burst occurs during pre- and initial swing. During fast walking, the RF contraction helps to accelerate the shank in initial swing, and the hamstrings act to decelerate it in terminal swing (Nene et al. 1999). Swing phase important kinematic events occur mainly in the sagittal plane. During initial swing, the peak of knee flexion (PKF), which accounts for the maximum degree of knee flexion during the gait cycle, compensates for ankle plantar flexion and helps with foot clearance (Anderson et al. 2004). Since the lower limb acts as a compound pendulum, hip and knee flexion occur with very little muscular intervention during normal swing phase (Piazza and Delp 1996). In mid-swing, this mechanism generates knee extension, whereas the hip keeps going into flexion. During terminal swing, hip flexion reaches its maximum and the knee is almost extended before initial contact. The only active muscle throughout swing phase is
Swing Phase Problems in Cerebral Palsy
1095
the tibialis anterior that brings the ankle into neutral position acting against gravity. The actions of hamstrings and tibialis anterior stabilize the knee and ankle positions for landing. In the transverse and coronal planes, pelvic internal rotation and hip adduction in terminal swing provide adequate step length and proper foot orientation at heel contact. Hip abduction is observed during initial swing and helps with limb advancement. Abnormal kinematics during swing are mainly represented by insufficient or delayed peak of knee flexion (stiff-knee gait) and excessive ankle plantar flexion. Transverse plane abnormal kinematics are frequently seen in patients with hemiplegia and tend to be related to excessive pelvic rotation and foot deviations.
State of the Art Swing phase problems in cerebral palsy (CP) are mainly represented by limited knee flexion (stiff knee) and insufficient ankle dorsiflexion. An insufficient or delayed PKF and similarly excessive ankle plantar flexion during swing compromise foot clearance and adequate foot position at initial contact. Although terminal stance kinematics and kinetics influence swing phase events, spasticity of the RF and poor distal motor control constitute the main causes of these gait deviations. Stiff-knee management is addressed to diminish RF spasticity or to treat RF contractures. In children with CP abnormal RF activity is almost always present. Rectus femoris distal tendon transfer to knee flexor muscles has been advocated to avoid the knee extensor moment generated by RF abnormal activity during swing and to create a knee flexion moment. Studies based on cine phase-contrast magnetic resonance imaging (MRI) showed that the RF is not converted to a knee flexor after its distal tendon is transferred to the posterior side of the knee, but its capacity for knee extension is diminished by the surgery. Also, the three-dimensional reconstruction (MRI) of RF muscle geometry after the transfer demonstrated that the muscle didn’t follow a straight course from its origin to its new insertion, but a sharp deviation was frequently seen (Asakawa et al. 2004). In addition, scar tissue between RF and the underlying muscles was visible and could be the cause of relapses after RF transfer. Long-term follow-up studies showed variable results regarding improvement of the magnitude of PKF, but improvements in timing of knee flexion and overall knee range of motion remained stable over time (Dreher et al. 2012; Thawrani et al. 2012). Distal RF tendon resection has been proposed as an alternative to the transfer. A short-term follow-up study showed that RF distal tendon resection was associated with improved knee range of motion and timing of PKF in swing, and the absolute values of PKF became normal for those patients who showed abnormal preoperative values (Presedo et al. 2012). Ankle excessive plantar flexion and dynamic varus deviations are commonly observed in patients with spastic hemiplegia (Gage 1991). Botulinum toxin and orthoses are currently used to diminish spastic plantar flexion and varus deviations and to improve foot clearance in patients with ankle dorsiflexor weakness. Surgical muscular-tendon lengthenings may be indicated to treat fixed contractures. Davids
1096
A. Presedo
et al. (2011) demonstrated an improvement in ankle dorsiflexion following ankle plantar flexor surgery in selected children with CP. They explained this by the fact that correction of a fixed constraining equinus contracture of the ankle plantar flexors may unmask preexisting ankle dorsiflexion function. Split anterior tibial tendon transfer, combined with gastrocnemius release and/or tibialis posterior lengthening, has been advocated to treat spastic equinovarus deformity. Hoffer et al. (1985) demonstrated that this procedure was reliable and effective to correct dynamic varus deviations in patients with CP, after a follow-up of 10 years. Other softtissue procedures like tibialis posterior transfer to the forefoot tend to be abandoned in patient with spasticity because of danger of overcorrection and resultant calcaneus-valgus deformity. Bony procedures (calcaneal osteotomies and triple arthrodesis) may be indicated to correct fixed deformities.
Swing Phase Abnormal Kinematic Patterns and Causes of Pathological Function Computerized gait analysis has become the gold standard for the evaluation of complex gait problems in patients with cerebral palsy (Narayanan 2007; Gough and Shortland 2008). Based on the kinematic study of lower limb motion, common abnormal patters for the individual joints have been described. The recognition of these patterns has conducted to the elaboration of gait classifications (Rodda and Graham 2001). Without any intervention, these patterns of gait are extremely predictable over time. Although therapy decisions are particular to each patient, the knowledge of these abnormal patterns should allow for a better comprehension of pathology and conduct to a more appropriate treatment.
Hip and Pelvis Pathology The hip and pelvis play a major role during gait, ensuring power generation and balance of the upper body segment. Hip extensor muscles provide 50% of power during normal walking (the other 50% coming from ankle plantar flexors), and in patients with CP, since motor control is somehow preserved around the hip, compensatory mechanisms occur mainly at this level (Winter 1991; Õunpuu 1994). During swing phase, pelvic internal rotation and hip adduction in terminal swing provide adequate step length and proper foot orientation at heel contact. Hip abduction is observed during initial swing and helps with limb advancement. Excessive hip flexion throughout the cycle is commonly seen in patients with severe involvement (GMFCS III–IV). The typical kinematic pattern called “doublebump pelvic pattern” (Fig. 1) is characterized by an increased anterior pelvic tilt in stance phase and a second one in swing phase (Gage 1991). This abnormal pattern is related to a poor dissociation between pelvic and hip motion. Anterior pelvic tilt peak occurs always at single-limb stance under the influence of spastic hip flexors and/or weak hip extensors. “Single-bump pelvic pattern” is seen in patients with
Swing Phase Problems in Cerebral Palsy
1097 Pelvic Tilt
30.0
Angle (degrees)
Ant
25.0
20.0
15.0
10.0
5.0
Post
0.0 −5.0
0
10
20
30
40
50
60
70
80
90
100
Angle (degrees)
Fig. 1 The green line represents normal pelvis kinematics in the sagittal plane. The blue and red lines represent abnormal “double-bump pattern” kinematics for the right and left lower limb, respectively
hemiplegia (Fig. 2). Anterior pelvic tilt peak occurs in terminal stance, only on the hemiplegic side, and pelvis returns to normal alignment in swing phase as the stretch of the hip flexors ends when the hip moves into flexion. Anterior pelvic tilt throughout the gait cycle is commonly seen in young children with spastic diplegia who walk with increased cadence to compensate for the lack of stability related to spasticity of ankle plantar flexors. In this case, the shank does not advance normally over the foot, and so the trunk and pelvis bend forward in order to move the center of mass along the line of progression. Hip and pelvic coronal and transverse plan abnormal patterns present often combined. In diplegic and hemiplegic patients with distal involvement, these patterns may represent coping responses. In hemiplegic patients, abnormal rotation of pelvis and hip are very common and may be related to asymmetric neurological involvement or to torsional bony deviations. Internal hip rotation on the hemiplegic side is compensated by external rotation of the ipsilateral hemipelvis (Fig. 3). Increased pelvic rotation can be a compensation for reduced sagittal plane motion, in order to increase step length. During swing, hip abduction or circumduction represents a typical compensation for reduced sagittal plane motion. As a consequence of increased pelvic rotation, coronal hip motion will also increase. Increased
1098
A. Presedo Pelvic Tilt
Ant
30.0
Angle (degrees)
20.0
10.0
Post
0.0
0
10
20
30
40
50
60
70
80
90
100
Normalised (percent)
Fig. 2 In a patient with right hemiplegia, a “single-bump” pattern of pelvis kinematics is typically observed
coronal pelvic motion may be seen to compensate for a reduced hip abductor moment in stance (hip abductor insufficiency). In this case, the patient will lean the trunk laterally to shift the body center of mass and bring the GRF vector close to the hip center of rotation. Kinematic curves will show ipsilateral pelvic depression in stance and elevation in swing with decreased hip adduction in mid-stance and increased hip adduction in terminal stance and initial swing. In quadriplegic and diplegic patients with more severe involvement, hip dysfunction is commonly related to excessive femoral anteversion and adductor muscle contractures.
Knee Pathology The peak of knee flexion is generated by the action of ankle plantar flexors in terminal stance and by the hip flexors in pre-swing and initial swing, so the lower limb acts as a compound pendulum (Anderson et al. 2004). During initial swing, additional knee flexion is performed by the short head of biceps femoris to facilitate foot clearance (Perry 1992). The magnitude of the motion and the timing are critical to achieve the necessary knee flexion to clear the foot. The isometric contractions of RF and hamstrings during swing phase act to regulate knee flexion/extension in order to adjust to changes in walking speed (Nene et al. 1999). Excessive knee flexion pattern is characterized by greater than normal knee flexion throughout the stance phase. Increased knee flexion is usually maintained
Swing Phase Problems in Cerebral Palsy
Add
a
1099 Ab/Adduction Hanche
30
20
Angle (degrees)
10
0
Abd
–10
–20
–30 0
b
10
20
30
20.0
60 40 50 Normalized (percent)
70
80
90
100
70
80
90
100
70
80
90
100
Obliquité bassin
up
15.0
Angle (degrees)
10.0 5.0 0.0 –5.0
Down
–10.0 –15.0 –20.0 0
Int
c
10
20
30
40 50 60 Normalized (percent) Rotation bassin
30
20
Angle (degrees)
10
0
–10
Ext
–20
–30 0
10
20
30
40
50
60
Normalized (percent)
Fig. 3 (a) Patient with right hemiplegia. Hip frontal plane kinematics show an asymmetric pattern, with the right side being abducted throughout the cycle. This pattern of hip motion is related to asymmetric pelvic frontal (b) and transverse (c) alignment
1100
A. Presedo
during swing phase, and global knee range of motion tends to be very limited (“crouch gait”). In cerebral palsy, this pattern is usually associated to hamstring spasticity or contracture but can be also secondary to plantar flexor or hip extensor weakness. Insufficient knee flexion in swing phase has been defined as diminished and/or delayed peak of knee flexion and is referred to as “stiff-knee gait” (Sutherland and Davids 1993). Stiff-knee gait was first described as one of the gait classification patterns; however, this pattern of gait is not considered as a separate entity by Rodda and Graham classification (Rodda and Graham 2001) since knee stiffness can be part of different types of gait. Stiff-knee gait is frequent in patients with jump and crouch gait patterns and also in hemiplegic patients with knee involvement. This kinematic pattern has been initially described as a deficit of knee flexion during swing (Fig. 4). However, since patients who walk with a permanent knee flexion throughout the
Fig. 4 (a) This patient shows knee stiffness and difficulty to clear the foot during swing phase. (b) Kinematics curve shows insufficient peak knee flexion during initial swing with decreased knee range of motion. (c) There is a premature and prolonged activity of the rectus femoris muscle during terminal stance and swing phases
Swing Phase Problems in Cerebral Palsy
1101
cycle may have normal amount and timing of PKF but very limited knee range of motion that may be difficult to clear the foot, decreased knee range of motion is also considered one of the criteria to define stiff-knee gait (Moreau and Tinsley 2005). In children with CP, stiff-knee gait is almost always associated to abnormal activity of the rectus femoris during swing. Abnormal knee function is almost always present in children with spastic diplegia. In patients with spasticity, selective distal motor control is impaired, and so they are stance stability and propulsion. Young patients with mild involvement can compensate for this with an increase of cadence, so they preserve speed and swing knee flexion. However, propulsion tends to deteriorate over time and so it does swing knee function. Because of the lack of distal motor control, biarticular muscles (rectus femoris and hamstrings) tend to favor proximal function and act as hip flexors/extensors rather than control knee motion. It is common in diplegic and hemiplegic patients to show prolonged activity of RF and hamstring co-contraction during swing phase. Thus, both muscles are active through mid-swing when they should be silent. The RF acts as a primary hip flexor and secondarily extends the knee, reducing the peak knee flexion, and the hamstrings act primarily as hip extensors and not being able to counterbalance the extension caused by the rectus (Gage et al. 1987; Perry 1987; Sutherland et al. 1990). Given the pathology of these muscles, the ability to allow variation of cadence can be largely compromised in patients with CP. In addition to that, poor propulsion and deficit of passive hip flexion that results are frequent in patients with slow walks and also contribute to knee stiffness.
Ankle and Foot Pathology In patients with CP, there is almost always a dominance of the triceps over the ankle dorsiflexors (Gage 1991). During swing phase, the shape of the ankle kinematic curve can be normal or it can lack dorsiflexion (Fig. 5a). In patients with spastic diplegia, foot position relative to the shank is usually normal, and forefoot initial contact is commonly related to abnormal knee flexion rather to ankle plantar flexion. The gastrocnemius is usually spastic and can develop some degree of contraction, but the soleus tends to maintain normal length, so fixed plantar flexion due to triceps contracture is relatively uncommon in these children. During swing phase, although the dorsiflexors may still be overwhelmed by the plantar flexors, some degree of dorsiflexion is in general possible, and thus, ankle kinematics tends to occur within normal limits. In children with hemiplegia, contracture is common in both the soleus and gastrocnemius. As a result of the overactivity of the triceps and the dominance of the tibialis anterior and posterior over the peroneals and toe extensors, the foot is typically postured in equinovarus. This deformity tends to become rigid over time. Thus, patients with hemiplegia may show a permanent plantar flexion kinematic pattern (Fig. 5b). During swing phase, problems with foot clearance and inadequate foot position in terminal swing can result from abnormal ankle kinematics.
1102
A. Presedo
Fig. 5 (a) Ankle kinematics in a spastic diplegic patient is characterized by plantar flexion at initial contact, followed by limited dorsiflexion in stance and insufficient dorsiflexion in swing phase. (b) Ankle kinematics in a hemiplegic patient with gastrocsoleus contracture shows permanent plantar flexion
Swing Phase Problems in Cerebral Palsy
1103
Management of Swing Phase Problems in Cerebral Palsy Treatment strategies should be based on the age of the patient and the severity of neurologic involvement. Primary problems (spasticity) should receive specific treatment with the aim of prevent secondary problems (muscular contractures and bone deformities). If orthopedic surgery procedures are needed, these should be planned within a multidisciplinary team, taking into consideration the whole child and not just his motor-skeletal parts.
Excessive Hip Adduction In diplegic and hemiplegic patients with distal involvement, abnormal hip and pelvis kinematic patterns may represent coping responses. In hemiplegic patients, abnormal rotation of pelvis and hip are very common and may be related to asymmetric neurological involvement and/or to torsional bony deviations. Increased hip adduction in terminal stance and initial swing is commonly seen in diplegic patients with more severe involvement and tends to be related to excessive femoral anteversion and adductor muscle contractures. Excessive hip adduction in swing can compromise foot clearance and cause tripping and falling. Therapeutic goals are oriented to treat adductor muscle contractures and bony torsions.
Stiff-Knee Gait Stiff-knee gait management in children with cerebral palsy will focus on treatment of RF spasticity and/or contractures. Rectus femoris surgical procedures are often performed as part of multilevel surgery; therefore, improvement on stance phase parameters can largely contribute to a better knee flexion during swing. Adequate knee flexion velocity at toe-off has been proved to be the most important kinematic factor to achieve normal swing phase knee flexion (Goldberg et al. 2004). Hip flexion moment, generated by the iliopsoas muscle, and ankle plantar flexion, generated by the gastrocnemius, were identified as the parameters that contribute most to increasing knee velocity during double support. Since prolongation of phasic, late stance, RF activity into swing phase constitutes an electromyographic (EMG) pattern commonly associated with stiff-knee gait, RF proximal or distal release procedures have been proposed to improve knee flexion in patients with abnormal RF activity in swing phase (Sutherland et al. 1990). According to these authors, RF proximal release did not influence pelvic alignment or hip range of motion. They concluded that RF release would reduce the extensor properties of the muscle and thus facilitate passive knee flexion in swing. Since patients with cerebral palsy tend to walk slower than normal, hip flexion moment in pre-swing is often diminished. For this reason, Perry suggested to abandon proximal RF release and transfer the RF posterior to the axis of knee to enhance active knee flexion (Perry 1987). Indications for transfer included (a) diminished range of knee
1104
A. Presedo
flexion during swing phase, (b) excessive RF EMG activity during swing, and (c) a positive Duncan-Ely test. Õunpuu et al. (1993a) compared the outcomes of RF release versus transfer. Although the differences were not significant in terms of amount of variation in PKF and time to PKF in swing, this study showed that there was a tendency toward more normal values after the transfer. This tendency could be explained by the notion that the RF, transferred posterior to the axis of knee flexion, would generate a knee flexion moment in swing. The choice of the transfer site did not influence PKF, knee ROM, or transverse plan kinematics values (Õunpuu et al. 1993b). Hemo et al. (2007) compared the outcomes of two different techniques: RF distal release and RF transposition to the iliotibial band. They found similar improvement in knee ROM, PKF, and time to PKF after 1-year follow-up. Factors related to good results after RF transfer have been indicated: (a) preoperative RF prolonged activity limited to swing phase (Miller et al. 1997), (b) preoperative positive Duncan-Ely test (Kay et al. 2004), and (c) postoperative increase in knee flexion velocity at toe-off (Goldberg et al. 2006). More recent studies attempted to explain the action of the RF muscle following distal tendon transfer. Riewald and Delp (1997) investigated whether the RF converts to a knee flexor after being transferred to the semitendinosus muscle or to the iliotibial band. Rectus femoris EMG activity showed that the muscle generated an extensor moment in all of their subjects. Based on cine phase-contrast MRI, Asakawa et al. (2002, 2004) examined RF motion in vivo and muscle geometry after tendon transfer surgery. In the tendon transfer group, the RF moved in the direction of the knee extensors, and fiber excursions were reduced compared to vastus intermedius. These authors concluded that the RF was not converted to a knee flexor after its distal tendon was transferred to the posterior side of the knee, but its capacity for knee extension was diminished by the surgery. They suggested that scar tissue could form after RF transfer, making the RF adhere to the underlying muscles. They also examined three-dimensional MRI of patients who had a RF transfer and observed abnormal, low-signal intensity images that could represent scar tissue between the transferred muscle and the underlying vastii in each of the patients. Tridimensional models showed that the transferred muscles followed angular, deviated paths to their new insertions, suggesting that RF tendons were probably constrained by adhesions to the underlying muscles. Following these results, we believed there was insufficient evidence supporting transferring the RF, rather than performing a distal release. We also thought a complete tendon resection would reduce adherences and prevent relapses (Fig. 6). We therefore determined to assess the outcome of children with spastic diplegia following RF distal tendon resection as a part of multilevel surgery, in order to (a) evaluate the improvement of knee ROM during gait, (b) measure changes in maximum knee flexion and time to PKF during swing phase, and (c) compare our results with those reported in the literature after RF distal release or transfer, since we are not aware of any previous reports on RF distal tendon resection. Our results, after 2-year follow-up, showed that RF distal tendon resection was associated with improved knee ROM and timing of peak knee flexion in swing, and the absolute values of peak knee flexion became normal for those patients who showed abnormal preoperative values (Presedo et al. 2012).
Swing Phase Problems in Cerebral Palsy
1105
Fig. 6 Rectus femoris distal tendon resection. The tendon is transected at the muscular junction and removed completely in order to avoid subsequent adherences to the underlying muscles
Long-term follow-up studies showed variable results in terms of PKF magnitude and timing but agreement on knee range of motion improvement that was maintained at final follow-up. Moreau and Tinsley (2005) compared the outcomes of a group of patients who had RF transfer as part of multilevel surgical procedures to another group of patients who did not undergo any RF procedure. They found the RF transfer helped to maintain knee ROM and PKF values over time, after a minimum follow-up of 3 years. Dreher et al. (2012) found a significant improvement in timing of PKF, knee range of motion, and knee flexion velocity after 9 years follow-up. Thawrani et al. (2012) reported an improvement in PKF magnitude and timing after 7-year follow-up.
Insufficient Ankle Dorsiflexion and Foot Varus Deviation Sagittal and coronal plan deviations of the ankle and foot are common in hemiplegic patients. These patients may show a permanent plantar flexion pattern, combined with hind foot varus deviation. A gait classification based on kinematic abnormal patterns in the sagittal plan is commonly used for hemiplegic patients (Winters et al. 1987). Type 1 hemiplegia is characterized by a “drop foot” during swing phase. This is due to a lack of control of ankle dorsiflexors. There is not plantar flexor contracture although they tend to be spastic. Because of the absence of muscle contracture, ankle dorsiflexion is normal in mid-stance, but the ankle is in plantar flexion at the initial contact. The treatment of this type of pattern is a leaf spring or an articulated ankle foot orthosis (AFO). Spasticity treatment can be indicated in order to improve ankle dorsiflexion during swing in patients with relative good motor control and active dorsiflexion in physical exam. Type 2 hemiplegia is characterized by a permanent ankle plantar flexion, during stance and swing phases. There is spasticity and/or contracture of plantar flexors. The management of this type of gait is based on spasticity treatment by botulinum
1106
A. Presedo
toxin injections and/or muscular lengthenings, if necessary. If there is a mild contracture, spasticity management combined with walking casts can be very effective to improve ankle range of motion. We tend to use knee immobilizers during nighttime in order to maintain the gastrocnemius in a stretched position. In case of more severe contracture, gastrocnemius lengthening or tendo-Achilles lengthening will be indicated. Most of the children will require the use of orthoses. Leaf spring orthoses will be adequate in absence of knee recurvatum; however, if the knee is fully extended or in recurvatum, articulated AFO with limitation of plantar flexion may be more convenient. Most of the diplegic and some hemiplegic patients with proximal involvement (type 4 hemiplegia) show combined coronal and transverse plane problems. In the coronal plane, excessive hip adduction secondary to adductor spasticity or contracture and hip subluxation represent the main issues. In the transverse plane, pelvic and hip abnormal rotations, femoral anteversion, tibial torsion, and foot deviations are commonly observed. Whereas management of sagittal plane problems is mainly address to spasticity and muscle contracture, bony surgery is often required to treat coronal and transverse plane abnormalities. Patients with type 4 hemiplegia present proximal involvement and a pattern of sagittal kinematics that can be similar to that seen on diplegic patients. As a result of the triceps overactivity and the dominance of the tibialis anterior and posterior over the peroneals, the foot typically in equinovarus tends to become rigid over time. During swing phase, due to spasticity or contracture of tibialis posterior and the predominance of tibialis anterior and extensor hallucis longus, the foot is deviated in varus and supination. In patients with severe spasticity and/or contracture of hip flexors and adductors, hip subluxation is not uncommon. Management of this type of hemiplegia may include spasticity treatment, but more often, muscular lengthening is required. Split anterior tibial tendon transfer, combined with gastrocsoleus and tibialis posterior lengthening, is recommended to correct dynamic equinovarus foot deviation (Hoffer et al. 1985; Davids et al. 2011). Bony surgery to address foot deformities and femoral torsion can be also part of the surgical program. After surgery, the most common type of orthoses used is a solid AFO that will help with knee extension and will also avoid ankle plantar flexion during swing. With time, most of the patients tend to keep a leaf spring AFO to help with clearance.
Summary Swing phase abnormal kinematics can compromise foot clearance and adequate limb pre-positioning for initial contact. In children with cerebral palsy, these problems are related to spasticity, muscular weakness, and poor motor control. Stiff-knee gait represents the most common swing phase kinematic problem in these patients. Although knee flexion during swing is related to terminal stance and pre-swing kinematics, rectus femoris prolonged activity during swing is almost always present, and stiff-knee treatment is oriented to diminish the extensor effect of this muscle. Long-term studies have demonstrated that after rectus femoris surgery (distal tendon
Swing Phase Problems in Cerebral Palsy
1107
resection or transfer), swing phase knee kinematics and knee range of motion tend to improve, and this improvement lasts over time. Difficulties with foot clearance caused by ankle dorsiflexor weakness, tendo-Achilles contracture, or hindfoot varus deviation are frequent in hemiplegic patients. Treatment focuses on spasticity management, musculotendinous releases, tibialis anterior transfer, or bony procedures to correct fixed deformities.
References Anderson FC, Goldberg SR, Pandy MG, Delp SL (2004) Contributions of muscle forces and toe-off kinematics to peak knee flexion during the swing phase of normal gait: an induced position analysis. J Biomech 37:731–737 Annaswamy TM, Giddings CJ, Della Croce U, Kerrigan DC (1999) Rectus femoris: its role in normal gait. Arch Phys Med Rehabil 80:930–934 Asakawa DS, Blemker SS, Gold GE, Delp SL (2002) In vivo motion of the rectus femoris muscle after tendon transfer surgery. J Biomech 35:1029–1037 Asakawa DS, Blemker SS, Rab GT, Bagley A, Delp SL (2004) Three dimensional muscle-tendon geometry after rectus femoris tendon transfer. J Bone Joint Surg 86:348–354 Davids JR, Rogozinski BM, Hardin JW, Davis RB (2011) Ankle dorsiflexion function after plantar flexor surgery in children with cerebral palsy. J Bone Joint Surg Am 93:e 138, 1–7 Dreher T, Wolf SI, Maier M, Hagmann S, Vegvari D, Gantz S, Heitzmann D, Wenz W, Braatz F (2012) Long-term results after distal rectus femoris transfer as a part of multilevel surgery for the correction of stiff-knee gait in spastic diplegic cerebral palsy. J Bone Joint Surg Am 94:e 142, 1–10 Gage JR (1991) Gait analysis in cerebral palsy. Mac Keith Press, London, pp 61–95 Gage JR, Perry J, Hicks RR, Koop S, Werntz JR (1987) Rectus femoris transfer to improve knee function of children with cerebral palsy. Dev Med Child Neurol 29:159–166 Goldberg SR, Anderson FC, Pandy MG, Delp SL (2004) Muscles that influence knee flexion velocity in double support: implications for stiff-knee gait. J Biomech 37:1189–1196 Goldberg SR, Õunpuu S, Arnold AS, Gage JR, Delp SL (2006) Kinematic and kinetic factors that correlate with improved knee flexion following treatment for stiff-knee gait. J Biomech 39:689–98. Gough M, Shortland AP (2008) Can clinical gait analysis guide the management of ambulant children with bilateral spastic cerebral palsy? J Pediatr Orthop 28:879–883 Hemo Y, Aiona MD, Pierce RA, Dorociack R, Sussman M (2007) Comparison of rectus femoris transposition with traditional transfer for the treatment of stiff knee gait in patients with cerebral palsy. J Child Orthop 1:37–41 Hoffer MM, Barakat G, Koffman M (1985) 10-year follow-up of split tibial tendon transfer in cerebral palsied patients with spastic equinus deformity. J Pediatr Orthop 5:432–434 Kay RM, Rethlefsen SA, Kelly JP, Wren TAL (2004) Predictive value of the Duncan-Ely test in distal rectus femoris transfer. J Pediatr Orthop 24:59–62 Miller F, Dias R, Lipton GE, Albarracin JP, Dabney KW, Castagno P (1997) The effect of rectus EMG patterns on the outcome of rectus femoris transfers. J Pediatr Orthop 17(5):603–607 Moreau N, Tinsley S (2005) Progression of knee joint kinematics in children with cerebral palsy with and without rectus femoris transfers: a long-term follow up. Gait Posture 22:132–137 Narayanan UG (2007) The role of gait analysis on the orthopaedic management of ambulatory cerebral palsy. Curr Opin Pediatr 19:38–43 Nene A, Mayagoitia R, Veltink P (1999) Assessment of rectus femoris function during initial swing phase. Gait Posture 9:1–9 Õunpuu S (1994) The biomechanics of walking and running. Clin Sports Med 13:843–863
1108
A. Presedo
Õunpuu S, Muik E, Davis RB 3rd, Gage JR, DeLuca PA (1993a) Rectus femoris surgery in children with cerebral palsy. Part I. The effect of rectus femoris transfer location on knee motion. J Pediatr Orthop 13:325–330 Õunpuu S, Muik E, Davis RB 3rd, Gage JR, DeLuca PA (1993b) Rectus femoris surgery in children with cerebral palsy. Part II. A comparison between the effect of transfer and release of the distal rectus femoris on knee motion. J Pediatr Orthop 13:331–335 Perry J (1987) Distal rectus femoris transfer. Dev Med Child Neurol 29:53–58 Perry J (1992) Gait analysis: normal and pathological function. Slack, Thorofare, pp 1–19 Piazza SJ, Delp SL (1996) The influence of muscles on knee flexion during the swing phase of gait. J Biomech 29:723–733 Presedo A, Megrot F, Ilharreborde B, Mazda K, Penneçot GF (2012) Rectus femoris distal tendon resection improves knee motion in patients with spastic diplegia. Clin Orthop Relat Res 470(5):1312–1319 Riewald SA, Delp SL (1997) The action of the rectus femoris muscle following distal tendon transfer: does it generate knee flexion moment? Dev Med Child Neurol 39:99–105 Rodda J, Graham HK (2001) Classification of gait patterns in spastic hemiplegia and spastic diplegia: a basis for a management algorithm. Eur J Neurol 8(Suppl 5):98–108 Sutherland DH, Davids JR (1993) Common gait abnormalities of the knee in cerebral palsy. Clin Orthop Relat Res 288:139–147 Sutherland DH, Santi M, Abel MF (1990) Treatment of stiff-knee gait in cerebral palsy: a comparison by gait analysis of distal rectus femoris transfer versus proximal rectus release. J Pediatr Orthop 10:433–441 Thawrani D, Haumont T, Church C, Holmes L, Dabney KW, Miller F (2012) Rectus femoris transfer improves stiff-knee gait in children with spastic cerebral palsy. Clin Orthop Relat Res 470(5):1303 Winter DA (1991) The biomechanics and motor control of human gait: normal, elderly and pathological, 2nd edn. University of Waterloo Press, Waterloo, pp 35–52, 75–85 Winters TF, Gage JR, Hicks R (1987) Gait patterns in spastic hemiplegia in children and young adults. J Bone Joint Surg Am 69:437–441
Strength Related Stance Phase Problems in Cerebral Palsy Justin Connor and Mutlu Cobanoglu
Abstract
Normal human gait results from a combination of several complex coordinated activities. The variable loss of control that is associated with cerebral palsy (CP) lesions can cause multiple interruptions in the gait cycle. To understand and properly evaluate pathologic gait patterns and to provide proper medical remedies for gait impairments due to CP, the clinician must understand the normal gait cycle. Instrumented gait analysis is a tool that provides detailed information and quantitative measurements throughout the gait cycle to evaluate individual gait patterns that help surgeons plan appropriate interventions. There are several prevailing abnormal gait patterns associated with spastic CP. These patterns are categorized into those affecting stance vs swing phases of gait. The patterns at the knee most often responsible for impeding the gait cycle by causing stance phase instability are crouch gait and back-kneeing in sagittal plane kinematic. Patterns such as back-kneeing cause knee hyperextension which is especially bad for delaying push-off and propelling forward; crouch knee gait causes dropping down into hip and knee flexion. These two types of gait patterns require different types of intervention due to differing muscle responses. Identification of the correct gait pattern in cerebral palsy is required to create the correct management algorithm.
J. Connor Nemours A.I. duPont Hospital for Children, Wilmington, DE, USA e-mail: [email protected] M. Cobanoglu (*) Department of Orthopedics and Traumatology, Adnan Menderes University Faculty of Medicine, Aydın, Turkey e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_54
1109
1110
J. Connor and M. Cobanoglu
Keywords
Cerebral palsy • Stance phase • Crouch gait • Back-knee • Ankle-foot orthosis • Knee-ankle-foot orthosis • Ground reaction force Abbreviations
AFO CP GRAFO HAT KAFO
Ankle-foot orthosis Cerebral palsy Ground reaction ankle-foot orthosis Head, arm, trunk Knee-ankle-foot orthosis
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Back-Kneeing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Crouch Knee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1110 1111 1112 1113 1113 1116 1118 1118 1118
Introduction Normal human gait is composed of a complex combination of coordinated motor activities which require precise control from the central nervous system. These motions control balance, motor activity, and cognitive decision-making. The gait cycle is divided into stance and swing phases. For gait to be productive and effective, there are five prerequisites as described by Perry: stability in stance, swing phase clearance, foot preposition in terminal swing, adequate step length, and energy conservation (Gage 1993). The role of stance phase (comprises 60% of the normal gait cycle) is to provide support for weight bearing and stability to provide power for push-off. Swing phase (comprises 40% of the normal gait cycle) advances the limb forward, thus allowing forward motion. These phases are supported by forces which are created at the ankle, knee, and hip joints by internal moments produced by attached muscles. The measurement of these forces at each joint through instrumented gait analysis allows for evaluation of kinetic moments. These kinetic measures create resultant vector forces by muscles on the bones of the body. Directly measuring muscle force is clinically impractical; however, net joint forces may be indirectly measured as the opposing ground reaction force. The residual of the ground reaction force at each joint has a direction and distance from the defined center of the joint that impacts gait. Concentric contraction of hip extensors, knee extensors, and ankle plantar flexors in stance phase maintain a ground reaction force in front of the knee joint during the stance phase, leading the erect posture during gait (Fig. 1) (Kadhim and Miller 2014).
Strength Related Stance Phase Problems in Cerebral Palsy
1111
Fig. 1 As shown in the figure, concentric contraction of hip extensors, knee extensors, and ankle plantar flexors in stance phase maintain a ground reaction force (red arrows) in front of the knee joint during the stance phase, leading the erect posture during gait
What is happening during stance phase? On the first contact to the ground, the knee is slightly flexed (approximately 5 ) to absorb the shock during initial heel strike. At this moment the quadriceps and hamstrings contract isometrically to stabilize the knee joint. Then the knee further flexes to 20 , shifting the center of gravity forward to allow advancement. After the short period of flexion, second half of stance starts. In this phase, the knee starts to extend with the activation of plantar flexion–knee extension couple: the soleus contracts and slows the progression of the tibia. This extension moment without activation of quadriceps muscle keeps the ground reaction force anterior to the knee. The hamstrings function in a closed chain to act as hip extensors. In the terminal phase of stance, limb acceleration occurs with the concentric contraction of the plantar flexors, providing power. This complex coordination of muscular activities is often difficult to maintain due to the neurologic insult associated with CP. Orthopedists must understand normal and abnormal gait cycle patterns and measurement techniques to be able to accurately evaluate pathologic gait and to improve function.
State of the Art Muscular spasticity and contractures leading to loss of functional abilities can be observed in ambulatory children with cerebral palsy resulting in pathologic gait patterns (Bell et al. 2002; Johnson et al. 1997). These primary motor deficits can lead to secondary biomechanical and growth problems such as muscle tightness, joint contractures, and secondary bony malformations. Rodda et al. described a classification based on the pelvis, hip, knee, and ankle position during stance to assist in understanding these abnormal gait patterns. According to this classification, there is a subset of patients in which the ankle is in equinus and knee and hip extend fully. Spasticity of the hamstrings and hip flexors increases gradually and leads to flexion of the knee and hip. This has been described as a jump gait. The classic crouch gait is manifested by ankle dorsiflexion with hip and knee flexion throughout the stance
1112
J. Connor and M. Cobanoglu
phase (Rodda et al. 2004). Increased body mass associated with the pubertal growth spurt may advance a child from a jump gait pattern to a crouch gait pattern (Rodda et al. 2006). The gait pattern also becomes asymmetric (Rodda et al. 2004). Crouch gait and back-kneeing (genu recurvatum) are the common types of gait patterns in ambulatory children with CP (Wren et al. 2005; Klotz et al. 2013). Both patterns affect stance phase stability by affecting pre-postponing the limb for proper push-off. The aim of this chapter is to investigate the properties and treatment results of these gait patterns.
Back-Kneeing The back-kneeing position is defined as hyperextension of the knee during the midstance phase and is frequently found in children with bilateral spastic cerebral palsy (Fig. 2). Eventually it may cause instability of the knee as the result of progressive stretching of the posterior knee capsule and ligamentous structures and over time cause secondary malformations of the tibial condyles as growth proceeds (Simon et al. 1978; Klotz et al. 2014). This condition can primarily occur as a result of three patterns. The first pattern is characterized by overactivity of the
Fig. 2 (a) The back-kneeing position is defined as hyperextension of the knee during the midstance phase of stance. (b) Ground reaction force in stance phase of the left lower limb is shown as red arrow which goes anterior to the knee axis. Average moments of lower limb joints in sagittal plan are shown. (c) Ten to forty percent of the gait cycle increased the internal hip extension moment seen because during the stance phase, the hip extensors try to compensate excessive hip flexion. This creates an increased internal extension moment. (d) Due to the severe back-kneeing during the stance phase, the knee flexor and capsule activate excessively to balance the high internal flexor moment caused by the hyperextension. (e) Because of persistent plantar flexion, the length of plantar flexors is shortened which limits the maximum internal plantar flexor moment needed after midstance (red line demonstrates right leg; blue line demonstrates left leg in c, d, and e)
Strength Related Stance Phase Problems in Cerebral Palsy
1113
gastrocsoleus muscles. The overactive gastrocsoleus causes the hyperextended knee through the coupling of ankle plantar flexion and knee extension. The second pattern is movement of the HAT (head, arm, trunk) center of gravity anterior to the knee in the presence of a weak gastrocnemius. The third pattern is movement of the HAT center of gravity posterior of the hip joint but anterior of the knee joint (Simon et al. 1978). Klotz et al. reported that equinus was a major underlying factor in primary back-knee gait (Klotz et al. 2014). Furthermore, this problem may occur secondarily after hamstring lengthening for the correction of flexed knee gait (Dreher et al. 2012).
Treatment The first consideration in the treatment of back-kneeing is to assess whether the gastrocnemius has adequate length to allow for dorsiflexion past neutral with knee extension. A hinged ankle-foot orthosis (AFO) with a 90 degree plantar flexion stop is the preferred orthosis for children whose back-kneeing results from overactivity of gastrocnemius spasticity. If dorsiflexion of the ankle with knee extension is possible, articulated AFO that allows 3–5 of dorsiflexion while limiting plantar flexion to minus 5 should be prescribed (Miller 2005). By setting the plantar flexion stop at 5 of dorsiflexion, these children will be forced into knee flexion in stance if they are independent ambulators. If the ground reaction force is moving either significantly in front or behind the knee in the presence of a plantar flexor weakness, a non-articulated solid ankle AFO should be used to control the ankle. If individuals have any shortness of the gastrocsoleus, the knee will hyperextend and go into back-kneeing. On the other hand, if individuals have plantar flexor weakness, back-kneeing can also occur due to increased ankle dorsiflexion and a HAT segment anterior to the knee. Instrumented gait analysis has shown an increased trunk lean in patients walking with assistive devices. In spite of appropriate orthotics, these assistive devices lead to an increased anterior trunk lean which may cause progressive back-kneeing and the development of pain (Simon et al. 1978; Klotz et al. 2014; Miller 2005). In the presence of this kind of progressive back-kneeing, the treatment method with AFO must be changed with knee-ankle-foot-orthosis (KAFO) with extension blocking hinges. But it is important to make sure that there is no contracture of the gastrosoleus. Ankle dorsiflexion has to be 5–10 in knee extension or the gastrocnemius should be lengthened (Dreher et al. 2012; Miller 2005).
Crouch Knee Crouch gait is common in ambulatory children with cerebral palsy (CP). The definition of a crouch gait is increased knee flexion in midstance with increased ankle dorsiflexion and usually increased hip flexion (Fig. 3). The crouch pattern may be seen in all levels of severity; however, it is primarily encountered in moderate and
1114
J. Connor and M. Cobanoglu
Fig. 3 (a) The definition of a crouch gait is increased knee flexion in midstance with increased ankle dorsiflexion and usually increased hip flexion. (b) Ground reaction force in stance phase of the lower limbs is shown as red arrow which goes posterior to the knee axis. Average moments of lower limb joints in sagittal plan are shown. (c) In this the gait cycle increased internal hip flexion moment seen at the end of stance phase in order to prepare the clearance during swing phase. (d) Due to the persistent knee flexion, knee extensors activate throughout the stance phase to balance the high knee internal extension moment caused by the knee flexion to prevent the possible collapse. (e) Because of persistent dorsiflexion, the plantar flexors activate to balance high ankle internal dorsiflexion moment; however, due to weakness of plantar flexors, the balance has to be maintained with strong quadriceps contraction (red line demonstrates right leg; blue line demonstrates left leg in c, d, and e)
severe diplegia. The ground reaction force is maintained close to the centers of the hip, knee, and ankle joints, reducing the demands on the antigravity support muscles. Failure of the total body extensor moment as a result of diminished ability of the hip, knee, or ankle plantar flexor moments may result in collapse of the extension posture into a flexion posture, described as crouch gait (Rodda et al. 2006). The rapid onset of weight and height growth may cause knee flexion contractures, hamstring contractures, deficient foot moment arm, and gastrocsoleus weakness which can contribute to the classic crouch gait. Knee flexion in stance phase increases and the foot starts to dorsiflex, and severe planovalgus foot deformities, characterized by heel equinus, talonavicular joint dislocation, and midfoot break during stance, develop and reduce lever arm (Kadhim and Miller 2014). The toe walking with knee flexion pattern (jump gait) is typically not seen in full adolescence or nearly adult-sized individuals. The muscles and joints are not strong enough to support the body weight for chronic ambulation with the typical early childhood toe-walking pattern. The other lever arm dysfunctions affecting stance stability are rotational deviations at the femurs, tibias, and feet. External tibial torsion reduces the extension capacity of the soleus and hip extensors, as it misdirects the forward propulsion vector, thus decreasing power, especially if the torsion is greater than 30 of normal (Hicks et al. 2007). Furthermore, the planovalgus foot is usually associated with
Strength Related Stance Phase Problems in Cerebral Palsy
1115
external tibial torsion. In the presence of gastrosoleus weakness in crouch gait, the ground reaction force creates a flexion moment on the knee and hip by passing through the knee posteriorly and the hip anteriorly. To prevent the knee from collapsing into flexion, quadriceps activity must increase during stance. The rectus femoris however not only acts as a hip extensor of the quadriceps but also contributes to hip flexion in a closed chain. Hip extensors are mechanically disadvantaged and the hamstrings cause increasing knee flexion. Over time, fixed hamstrings and knee flexion contracture, which are responsible for further exacerbation of crouch, may develop. The patellar tendon elongates gradually and extensor mechanism weakens. A secondary etiology for crouch may be a significant hip flexion contracture, which can limit knee extension in midstance. Although hamstring contracture is considered a primary cause for crouch, many patients have normal length hamstrings (Arnold et al. 2006; Delp et al. 1996). And also it has also been observed that excessive knee flexion typically accompanies excessive hip flexion due to shortened hip flexors throughout the gait cycle (Delp et al. 1996). The shortened hip flexors lead to anterior pelvic tilt facilitating hamstring tightness and causing knee flexion in stance phase (Kedem and Scher 2016). As a result of increasing crouch, the stress on the knee extensor mechanisms to support weight bearing may lead to the complaint of knee pain. Standing with >30 of knee flexion increases the forces acting on the quadriceps, patella, and proximal part of the tibia and requires the quadriceps muscle to work at >50% of its maximum moment-generating capacity in order to stabilize the knee joint (Perry et al. 1975). Tibial tubercle apophysitis or even chronic fractures through the distal pole of the patella may occur, especially during rapid growth. Progressive failure of the knee extensor mechanism is associated with gait deterioration, increased dependence on walkers or crutches, and the need for wheelchair use in the community (Rodda et al. 2006). As gait deteriorates the position of the feet may progress into increased. Orthotics lose their ability to support the collapsing feet, and ultimately the child may lose the ability and motivation to walk (Miller 2005). A child’s weight plays an important role in the evolution of gait patterns through maturity. It should be monitored on every clinic visit particularly throughout the pubertal growth spurt. During this period the child may exhibit new complaints of pain due to increasing stress on the knees or feet. Also, the physical examination should be monitored, being mindful of passive knee extension and popliteal angle measurements, to monitor progressive hamstring contractures or fixed knee flexion contractures that may impede gait. Objective measures to evaluate crouch gait through instrumented gait analysis include kinematic evaluation of the magnitude of the knee flexion in midstance, excessive dorsiflexion of the ankle, and the knee excessive internal knee extension moments in midstance. If the range of motion (ROM) of ankle is normal or below normal, and ROM does not increase with the knee flexion, the ankle weakness and foot moment arm are the most likely reasons of the crouch. If the knee extends to the limits of the fixed knee flexion contracture measured on physical examination, the knee joint contracture is most likely a cause. If the ankle has a high plantar
1116
J. Connor and M. Cobanoglu
flexion moment with a high knee flexion moment, a combination of contracture of the gastrocnemius and the hamstrings is most likely a cause. If the hip extension peak occurs early, and a significant hip flexion contracture is positive in the physical examination, hip flexion contracture may also contribute to knee flexion deformity in the midstance phase (Miller 2005).
Treatment Appropriate treatment for crouch gait should focus on early detection, by monitoring examinations every 6 months during middle childhood and intervention before the problem becomes severe. Mild crouch can be initially controlled conservatively, with spasticity management, physical therapy, and foot orthoses. But the role of physical therapy and strengthening exercises in the treatment of crouch is controversial. A meta-analysis of randomized trials concluded that strengthening interventions had no effect on strength, walking speed, or activity level in children with CP (Scianni et al. 2009), though there are reports that find that strengthening exercises may lead to some functional improvement in patients with crouch (Damiano et al. 1995, 2009). The aim of the interventions in these children is to reduce knee flexion and prevent gait deterioration ultimately maintaining gait efficiency and walking activity in daily life. Initially a solid AFO provides stability to the ankle and foot and provides a stable base of support for children to stand. These can usually be used at the preambulatory stage between the ages of 18 and 24 months. It is easy to done and works well for child less than 30 kg. As children gain better stability and start to walk using a walker, usually between the ages of 3 and 4 years, the ankle hinge can be added to allow dorsiflexion but limit plantar flexion. Most children who have good walking ability with diplegic and hemiplegic pattern involvement benefit from the transition to a hinged AFO at approximately 3 years of age. But hinged AFO is contraindicated if children have severe planovalgus or varus foot deformity and increased knee flexion in stance (Miller 2005). A rigid AFO can compensate for weakness of ankle plantar flexors and may normalize knee kinematics and kinetics effectively, but it has the disadvantage of inhibiting push-off power (Kerkum et al. 2015). Although AFO may affect ankle joint dorsiflexion during stance, ankle power and ankle plantar flexion moment, it does not affect proximal joint movement during gait (Rethlefsen et al. 1999). A ground reaction ankle-foot orthosis (GRAFO) can be used to enhance push-off power and is a commonly applied intervention in children with CP walking with crouch. A GRAFO features a solid pretibial shell to more effectively redirect the GRF vector and slow tibial progression as the center of pressure moves distally under the foot during stance (Kane et al. 2010). Extremities with knee and hip flexion contractures of 15 (Rogozinski et al. 2009). The GRAFO limits ankle dorsiflexion and reduces knee flexion in stance. This improved ankle and knee
Strength Related Stance Phase Problems in Cerebral Palsy
1117
position reduces the knee extensor moment in stance. The ankle can be brought to neutral dorsiflexion with the knee in full extension. If this is not possible, the orthosis cannot work and these children will require gastrocnemius and hamstring lengthening to accommodate the orthosis. The other point is that the foot-to-knee axis has to be in a relatively normal alignment which defines less than 20 of internal or external tibial torsion (Miller 2005). Transverse plane skeletal malalignments and foot segmental malalignments (planovalgus deformities) may lead to an increased external foot progression angle which moves the line of action of the ground reaction force lateral to the knee joint center. Recent studies utilizing dynamic computer models of the musculoskeletal system have demonstrated that increased external tibial torsion impedes the capacity of the soleus to extend the knee during stance, which supports the hypothesis of lever arm dysfunction as an important contributor to crouch gait (Hicks et al. 2007). As these children get heavier, this orthosis becomes more effective; however, it also has to be made stronger. As children approach 50–70 kg, the orthosis has to be constructed with a composite of carbon fiber or laminated copolymer to withstand the applied forces (Miller 2005). The surgery for crouch gait often includes many procedures at different joints. Single-event multilevel surgery is widely performed in cerebral palsy and focuses on improving gait function and pattern (Rodda et al. 2006). The preferred order is to start from proximal to distal. First the hip rotation should be corrected, with iliopsoas lengthening if needed. Distal hamstring lengthening if indicated should also be performed. Repeat hamstring lengthening may prevent or delay progressive crouch in patients with CP but does not result in long-term improvement in crouch gait (Rethlefsen et al. 2013). The hamstring lengthening followed by knee capsulotomy or femoral extension osteotomy (Novacheck et al. 2009) is performed if indicated. The surgery for foot deformity is done next; then an intraoperative assessment of the torsional alignment is used to make the final determination of the need for a tibial osteotomy. After the tibial osteotomy, the hip fully extends and the knee can be fully extended and lies in approximately 10 of external rotation. The foot-to-thigh alignment should be 20 external to neutral with neutral dorsiflexion (Kadhim and Miller 2014; Miller 2005). If the foot has a significant planovalgus or a midfoot break, it must be corrected. A stable and correctly aligned foot is mandatory in the correction of crouch because the ground reaction force has to be controlled through the foot as a functional moment arm (Kadhim and Miller 2014). Postoperative rehabilitation should start in the hospital with the goal of having children at least stand before discharge and plan for immediate home rehabilitation. Parents need to expect that the acute rehabilitation will take 3 months until these individuals are close to their preoperative function, and then it will take at least 1 year of rehabilitation to reach maximum function. If there is weakness or a tendency for the gastrocsoleus not to have good strength, a GRAFO has to be used postoperatively. This is the ideal time to use the articulated GRAFO, which will allow the gastrocsoleus to gain strength, and over 1–2 years, the orthotic can be weaned away and the correction will be maintained (Miller 2005).
1118
J. Connor and M. Cobanoglu
Conclusion There are a variety of factors that affect stance phase stability of the lower limb segment in gait. These factors may lead to progressive deterioration of gait and function and may ultimately lead to pain and disability. Through the use of routine monitoring of pubertal maturation, physical examination, and analysis of kinematics and moment kinetics through instrumented gait analysis, appropriate treatment plans can be acted upon to prevent progression of deformity and loss of function. Appropriate bracing adjuvants should also be utilized in addition to surgical interventions.
Cross-References ▶ Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy ▶ Natural History of Cerebral Palsy and Outcome Assessment ▶ Spasticity Effect in Cerebral Palsy Gait ▶ Swing Phase Problems in Cerebral Palsy
References Arnold AS, Liu MQ, Schwartz MH, Ounpuu S, Delp SL (2006) The role of estimating muscletendon lengths and velocities of the hamstrings in the evaluation and treatment of crouch gait. Gait Posture 23(3):273–81. Epub 2005 Jun 17 Bell KJ et al (2002) Natural progression of gait in children with cerebral palsy. J Pediatr Orthop 22(5):677–682 Damiano DL, Kelly LE, Vaughn CL (1995) Effects of quadriceps femoris muscle strengthening on crouch gait in children with spastic diplegia. Phys Ther 75(8):658–667. discussion 668–71 Damiano DL et al (2009) Can strength training predictably improve gait kinematics? A pilot study on the effects of hip and knee extensor strengthening on lower-extremity alignment in cerebral palsy. Phys Ther 90(2):269–279 Delp SL et al (1996) Hamstrings and psoas lengths during normal and crouch gait: implications for muscle-tendon surgery. J Orthop Res 14(1):144–151 Dreher T et al (2012) Development of knee function after hamstring lengthening as a part of multilevel surgery in children with spastic diplegia. J Bone Joint Surg 94(2). https://doi.org/ 10.2106/jbjs.j.00890 Gage JR (1993) Gait analysis. An essential tool in the treatment of cerebral palsy. Clin Orthop Relat Res 288:126–134 Hicks J et al (2007) The effect of excessive tibial torsion on the capacity of muscles to extend the hip and knee during single-limb stance. Gait Posture 26(4):546–552 Johnson DC, Damiano DL, Abel MF (1997) The evolution of gait in childhood and adolescent cerebral palsy. J Pediatr Orthop 17(3):392–396 Kadhim M, Miller F (2014) Crouch gait changes after planovalgus foot deformity correction in ambulatory children with cerebral palsy. Gait Posture 39(2):793–798 Kane K, Kyra K, John B (2010) Comparison of ground reaction and articulated ankle-foot orthoses in a child with lumbosacral myelomeningocele and tibial torsion. J Prosthetics Orthot 22(4): 222–229 Kedem P, Scher DM (2016) Evaluation and management of crouch gait. Curr Opin Pediatr 28(1): 55–59
Strength Related Stance Phase Problems in Cerebral Palsy
1119
Kerkum YL et al (2015) The effects of varying ankle foot orthosis stiffness on gait in children with spastic cerebral palsy who walk with excessive knee flexion. PLoS One 10(11):e0142878 Klotz MCM et al (2013) Reduction in primary genu recurvatum gait after aponeurotic calf muscle lengthening during multilevel surgery. Res Dev Disabil 34(11):3773–3780 Klotz MCM et al (2014) The association of equinus and primary genu recurvatum gait in cerebral palsy. Res Dev Disabil 35(6):1357–1363 Miller F (2005) Cerebral palsy. Springer Science & Business Media, New York Novacheck TF et al (2009) Distal femoral extension osteotomy and patellar tendon advancement to treat persistent crouch gait in cerebral palsy. Surgical technique. J Bone Joint Surg 91(Suppl 2): 271–286 Perry J, Antonelli D, Ford W (1975) Analysis of knee-joint forces during flexed-knee stance. J Bone Joint Surg 57(7):961–967 Rethlefsen S et al (1999) The effects of fixed and articulated ankle-foot orthoses on gait patterns in subjects with cerebral palsy. J Pediatr Orthop 19(4):470–474 Rethlefsen SA et al (2013) Repeat hamstring lengthening for crouch gait in children with cerebral palsy. J Pediatr Orthop 33(5):501–504 Rodda JM et al (2004) Sagittal gait patterns in spastic diplegia. J Bone Joint Surg 86(2):251–258 Rodda JM et al (2006) Correction of severe crouch gait in patients with spastic diplegia with use of multilevel orthopaedic surgery. J Bone Joint Surg 88(12):2653–2664 Rogozinski BM, Davids JR, Davis RB 3rd, Jameson GG, Blackhurst DW (2009) The efficacy of the floor-reaction ankle-foot orthosis in children with cerebral palsy. J Bone Joint Surg Am 91(10): 2440–2447 Scianni A et al (2009) Muscle strengthening is not effective in children and adolescents with cerebral palsy: a systematic review. Aust J Physiother 55(2):81–87 Simon SR et al (1978) Genu recurvatum in spastic cerebral palsy. Report on findings by gait analysis. J Bone Joint Surg Am 60(7):882–894 Wren TAL, Rethlefsen S, Kay RM (2005) Prevalence of specific gait abnormalities in children with cerebral palsy: influence of cerebral palsy subtype, age, and previous surgery. J Pediatr Orthop 25(1):79–83
Foot and Ankle Motion in Cerebral Palsy Jon R. Davids and Sean A. Tabaie
Abstract
There are three common ankle/foot segmental malalignment patterns seen in children with cerebral palsy (CP): equinus, equinoplanovalgus, and equinocavovarus. Each type of ankle/foot malalignment can be classified into three levels based upon the presence of dynamic muscle imbalance, fixed or myostatic deformity of the muscle-tendon unit, and fixed or skeletal deformities. Management of foot and ankle deformities in children with CP can consist of both nonsurgical (orthotics and pharmacotherapy) and surgical (muscle-tendon unit lengthening or transfer, skeletal osteotomies or arthrodeses) modalities. The goals of these surgeries are to rebalance muscle activity, restore range of motion, and realign the skeletal segments of the foot. Clinical decision-making for the management of the ankle/foot in children with CP involves a diagnostic matrix utilizing data from the clinical history, physical examination, radiographic imaging, and quantitative gait analysis. Keywords
Cerebral palsy • Foot • Ankle • Management
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Normal Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Disruption of Foot, Ankle Function, and Its Effect on Gait in Cerebral Palsy . . . . . . . . . . . . . . . Clinical Decision-Making: Management of Foot and Ankle Problems in CP . . . . . . . . . . . . . . . . Clinical History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Physical Examination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Radiographic Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1122 1122 1130 1133 1133 1134 1134
J.R. Davids (*) • S.A. Tabaie Northern California Shriner’s Hospital for Children, Sacramento, Sacramento, CA, USA e-mail: [email protected]; [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_58
1121
1122
J.R. Davids and S.A. Tabaie
Quantitative Gait Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Levels of Deformity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Treatment Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Surgical Treatment Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Surgical Treatment Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1135 1137 1138 1138 1140 1147 1148
Introduction Foot and ankle problems are common in children with cerebral palsy (CP). They affect both the ambulatory and nonambulatory children either by compromising the efficiency of gait or leading to issues with orthotic and shoe wear (Davids 2010). To approach the management of foot and ankle problems in a child with CP, it is important to understand normal gait and how it is disrupted in children with CP. This chapter will discuss normal gait at the foot and ankle and the disruption of normal gait in CP and then present the current paradigm for clinical decision-making for surgical management about the foot and ankle in children with CP.
Normal Gait The interaction between the ankle, foot, and the floor is the basis of normal gait. The function of the foot and ankle is based on a complex interplay between anatomy, physiology, and physics. Proper leads to efficient function of the knee and hip during gait (Davids et al. 2007, Inman 1969; Inman et al. 1981; Perry 1992). Standardized, consistent terminology should be used to describe the alignment of the separate segments of the ankle and foot (Ponseti et al. 1981). The understanding of foot and ankle function during normal gait is facilitated by considering the lower leg as consisting of four segments and the foot as three segments (Inman 1966; Inman et al. 1981; Perry 1992; Davids 2009a, 2010, 2016; Fig. 1): • • • •
Tibial or shank Hindfoot (talus and calcaneus) Midfoot (navicular, cuneiforms, and cuboid) Forefoot (metatarsals and phalanges) The foot should also be considered as consisting of two columns (Fig. 2):
• Medial column (talus, navicular, cuneiforms, great toe metatarsal, and phalanges) • Lateral column (calcaneus, cuboid, lesser toe metatarsals, and phalanges) Movement of the segments should be described as (Fig. 3):
Foot and Ankle Motion in Cerebral Palsy
1123
Fig. 1 Three segments of the foot. (a) Diagram of the AP view of the foot. The hindfoot segment is red, the midfoot segment is yellow, and the forefoot segment is white. (b) Diagram of the lateral view of the foot. The hindfoot segment is red, the midfoot segment is yellow, and the forefoot segment is white
Fig. 2 Two columns of the foot. (a) Diagram of the AP view of the foot. The medial column is red, and the lateral column is yellow. (b) Diagram of the lateral view (medial side) of the foot. The medial column is red. (c) Diagram of the lateral view (lateral side) of the foot. The lateral column is yellow
1124
J.R. Davids and S.A. Tabaie
Fig. 3 (a) Terminology for describing foot segmental alignment, utilizing the hindfoot as an example. When the plantar aspect of the hindfoot is deviated toward the body’s midline (red arrow), the segment is described as being inverted. When the plantar aspect is deviated away from the midline (yellow arrow), the segment is described as being everted. (b) When the distal aspect of the hindfoot is deviated toward the midline (red arrow), the segment is described as being adducted. When the distal aspect is deviated away from the midline (yellow arrow), the segment is described as being abducted
• Plantar aspect of segment – Inversion (toward the midline) – Eversion (away from the midline) • Distal portion of segment – Adduction (toward the midline) – Abduction (away from the midline) • Multi-planar motion – Supination (inversion and adduction) – Pronation (eversion and abduction) • Rotation of the segment about its longitudinal axis – External rotation (away from midline) – External rotation (away from midline) Function of the foot and ankle during the gait cycle is best appreciated by consideration of kinematics, kinetics, EMG (electromyography), and skeletal alignment (Fig. 4).
Foot and Ankle Motion in Cerebral Palsy
1125
Fig. 4 Illustration of kinematics, moment, and EMG throughout the gait cycle (IC initial contact, MSt midstance, TSt terminal stance, PSw preswing, ISw initial swing, MSw midswing, TSw terminal swing)
The interaction of the foot and ankle with the floor during the stance phase is best understood utilizing the concept of three rockers (Inman et al. 1981; Perry 1992; Davids 2009a, 2010). • First or heel rocker (Fig. 5) – Occurs from heel strike to foot flat during the loading response subphase of stance. – As the body progresses forward over the foot, the ground reaction force moves progressively distally through the foot, creating an external dorsiflexion moment. – The tibia advances forward over the foot, which is achieved by ankle dorsiflexion.
1126
J.R. Davids and S.A. Tabaie
Fig. 5 The first or heel rocker. The clinical photograph shows a heel strike at initial contact. The normal ankle kinematics, kinetics, and dynamic EMG during the gait cycle are plotted below the clinical photograph
35.0 Dorsiflexion Ankle Angle (degrees) 7.5 Plantar Flexion –20.0 2.0 Plantar Flexor Moment (Nm/kg)
0.8
Dorsiflexor –0.5 Gastrocnemius
Tibialis Anterior
0
25 50 75 % Gait Cycle
100
– This motion is controlled by eccentric activity of the ankle plantar flexor muscle group, which generates an internal plantar flexion moment. – This provides shock absorption during loading response. • Second or ankle rocker (Fig. 6) – Occurs as the tibia advances over the foot during the midstance subphase of stance. – With forward progression, the ground reaction force remains distal to the ankle joint, creating an increasing external dorsiflexion moment.
Foot and Ankle Motion in Cerebral Palsy
1127
Fig. 6 The second or ankle rocker. The clinical photograph shows the tibia advancing over the foot in midstance. The normal ankle kinematics, kinetics, and dynamic EMG during the gait cycle are plotted below the clinical photograph
35.0 Dorsiflexion Ankle Angle (degress) 7.5 Plantar Flexion –20.0 2.0 Plantar Flexor Moment (Nm/kg)
0.8
Dorsiflexor –0.5 Gastrocnemius
Tibialis Anterior
0
25 50 75 % Gait Cycle
100
– This motion is controlled by the concentric activity of the ankle plantar flexor muscle group, which continues to generate an internal plantar flexion moment. – This provides stability during midstance. • Third or forefoot rocker (Fig. 7) – Prior to the initial contact of the opposite foot, the heel of the reference foot rises off the ground, and dorsiflexion occurs through the metatarsophalangeal joints of the forefoot.
1128
J.R. Davids and S.A. Tabaie
Fig. 7 The third or forefoot rocker. The clinical photograph shows the heel rising and toes dorsiflexing in terminal stance. The normal ankle kinematics, kinetics, and dynamic EMG during the gait cycle are plotted below the clinical photograph
35.0 Dorsiflexion Ankle Angle (degress) 7.5 Plantar Flexion –20.0 2.0 Plantar Flexor Moment (Nm/kg)
0.8
Dorsiflexor –0.5 Gastrocnemius
Tibialis Anterior
0
25 50 75 % Gait Cycle
100
– The ankle starts to plantar flex as the reference limb is unloaded during preswing subphase of stance. – Largest moment generated by any single muscle group during the gait cycle is the internal plantar flexion moment generated by the plantar flexor group during third rocker in terminal stance. – This is essential for normal gait as it provides a rigid lever (i.e., stability) during terminal stance.
Foot and Ankle Motion in Cerebral Palsy
1129
Normal gait is the consequence of couple movements between the segments of the foot described earlier in this chapter. The alignments of the ankle and foot, and the activity of the ankle plantar flexor muscles, determine the location of the ground reaction force about the knee in mid and terminal stance which influences knee alignment. In general, ankle plantar flexion is associated with knee extension, and ankle dorsiflexion is associated with knee flexion. This biomechanical relationship is known as the plantar flexion/knee extension couple, and it is through this mechanism that control of the foot and ankle alignment can contribute to the control of knee alignment in stance phase (Inman et al. 1981; Inman 1966; Perry 1992; Davids 2009a; Fig. 8). In the swing phase of normal gait, the foot and ankle contribute to clearance and pre-positioning for the subsequent stance phase. • Initial swing – The combination of hip flexion (active), knee flexion (passive), and ankle dorsiflexion (active) serves to functionally shorten the limb to promote clearance. – The tibia or shank segment is externally rotating, and the ankle is plantarflexing.
Fig. 8 The plantar flexion/ knee extension couple. In the midstance of subphase of the gait cycle, increasing ankle plantar flexion (red arrow) shifts the ground reaction force (red arrow) anterior to the knee joint center, which causes the knee to extend. Increasing ankle dorsiflexion (black arrow) shifts the ground reaction force (black) behind the knee joint center, which causes the knee to flex
1130
J.R. Davids and S.A. Tabaie
• Midswing – The tibia or shank segment is rotating internally and the ankle is dorsiflexing. – Maximum shortening is achieved and clearance within 1 cm of the ground (proprioception) occurs. • Terminal swing – The coupled movements described in midswing continue, and the foot is maintained in a plantigrade position, perpendicular to the anatomical axis of the tibia or shank segment. This pre-positioning of the foot during terminal swing will result in a heel strike at the initial contact.
Disruption of Foot, Ankle Function, and Its Effect on Gait in Cerebral Palsy Foot and ankle deformities in children with CP are the consequence of a dynamic imbalance between the extrinsic muscles of the lower leg that control segmental foot and ankle alignment (Davids 2009b, 2010). The causes of imbalance are linked to: • Spasticity • Disrupted motor control • Impaired balance function Most often, the ankle plantar flexor muscles are overactive, and the ankle dorsiflexor muscles are ineffective. Additional imbalance patterns can be evident between the muscle groups that control foot and ankle supination and pronation. Ultimately, these muscle imbalances can lead to three common coupled foot and ankle segmental malalignment patterns in children with spastic-type CP. Other, more complex, or uncoupled segmental malalignment may occur but are much less common (Davids 2009b, 2010; Stevens 1988; Mosca 1995; Etnyre et al. 1993; Tylkowski et al. 2009). • Equinus deformity (Fig. 9) – Characterized by excessive plantar flexion of the hindfoot relative to the ankle. – In a purely equinus deformity, there is normal midfoot and forefoot alignment. • Equinoplanovalgus deformity (Fig. 10) – Characterized by equinus deformity of the hindfoot, along with pronation deformities of the midfoot and forefoot. – In this deformity pattern, the lateral column of the foot is shorter (anatomically and functionally) than the medial column. – Associated deformities of ankle and/or hallux valgus are commonly seen with equinoplanovalgus foot segmental malalignment. • Equinocavovarus deformity (Fig. 11) – Characterized by equinus deformity of the hindfoot, along with supination deformity of the midfoot and variable malalignment of the forefoot. – In this deformity pattern, the lateral column is functionally and/or anatomically longer than the medial column.
Foot and Ankle Motion in Cerebral Palsy
1131
Fig. 9 Plain radiographs of the foot in a child with equinus deformity. (a) AP view shows normal segmental alignment. (b) Lateral view shows hindfoot plantar flexion (diminished calcaneal pitch, normal is 25–30 ), with otherwise normal segmental alignment
Fig. 10 Plain radiographs of the foot in a child with equinoplanovalgus deformity. (a) AP view shows hindfoot pronation, talonavicular uncoverage, and forefoot abduction. (b) Lateral view shows hindfoot plantar flexion, midfoot pronation, and forefoot pronation
1132
J.R. Davids and S.A. Tabaie
Fig. 11 Plain radiographs of the foot in a child with equinovarus deformity. (a) AP view shows hindfoot supination and forefoot adduction. (b) Lateral view shows hindfoot varus, midfoot supination, and forefoot supination
– Associated deformities of the ankle may be seen with equinocavovarus segmental foot malalignment. The three common malalignment patterns described above are often correctable on manipulation in younger children with milder forms of CP. However, the foot and ankle deformities can become rigid and uncorrectable due to the following factors: • Increasing age and growth • Fixed shortening or myostatic deformities of the muscles • Development of permanent skeletal deformities in response to the malalignment pattern Foot and ankle segmental malalignment in children with CP can alter function during both the stance and swing phases of the gait cycle. The three segmental malalignment patterns are associated with a variety of gait disruptions and deviations (Davids 2010). • Common alterations to the normal gait cycle seen with all three segmental malalignment patterns: – Lack of heel strike at initial contact, disrupting the first rocker and shock absorption function in the loading response of the gait cycle – Compromised ankle plantar flexor muscles ability to generate an adequate internal plantar flexor moment during the second and third rockers
Foot and Ankle Motion in Cerebral Palsy
1133
– Inhibition of ankle dorsiflexion in swing phase, compromising clearance in midswing and proper positioning of the foot and ankle in terminal swing • Common alterations to the normal gait cycle seen with equinus and equinocavovarus malalignment patterns: – Disruption of the second rocker by blocking ankle dorsiflexion, thereby compromising stability in midstance. – Due to the hindfoot malalignment with these two deformity patterns, the lengths of the plantar flexor muscles are shortened, and their ability to generate tension is adversely affected. • Common alterations of the normal gait cycle with equinoplanovalgus segmental malalignment: – In this deformity pattern, the midfoot and forefoot segments remain in an unlocked alignment leading to excessive loading of the plantar, medial portion of the midfoot. – The malalignment of the midfoot and forefoot further compromises the momentgenerating capacity of the ankle plantar flexor muscles and effectively shortens the lever arm available to this muscle group during the third rocker. – External tibial torsion seen with equinoplanovalgus malalignment may contribute to an external foot progression angle, further affecting the lever arm available to the ankle plantar flexor muscles in terminal stance.
Clinical Decision-Making: Management of Foot and Ankle Problems in CP The management of foot and ankle deformities in CP can be very challenging. In order to provide the best care for this patient population, it is best to utilize a diagnostic matrix that incorporates the following (Davids 2009b, 2010; Davids et al. 2005, 2007; Inman et al. 1981; Perry 1992): • • • •
Clinical history Physical examination including observational gait analysis Plain radiographs Quantitative gait analysis
Clinical History Often patients present with specific complaints of gait abnormalities, such as pain with ambulation, tripping when ambulating, and/or issues with shoe wear. It is imperative to inquire and document the specifics of the symptoms, inciting and
1134
J.R. Davids and S.A. Tabaie
alleviating factors, duration of symptoms and their progression over time, and finally the treatment history to date.
Physical Examination The sequence of the physical exam in children with CP who have foot and ankle deformities should begin with a thorough inspection and progress to a hands-on evaluation of the foot segmental alignment in both weight-bearing and non-weightbearing conditions. Inspection of the plantar aspect of the feet should be done to evaluate for the presence of excessive or inadequate skin callous formation that could be due to abnormal loading patterns and/or problems with shoe/orthotic wear. Manual examination is performed to determine and document: • • • • •
Static standing alignment – assessed from the front, back, and both sides of the patient Intra- and intersegmental flexibility Rotational profile from the hips to the foot Range of motion – passive and active Muscle strength and selective motor control
The final key portion of the physical exam is the observational gait analysis. Observing ambulation needs to be done from multiple viewpoints in both the coronal and sagittal planes. The patient should be barefoot and wearing tight cycling style shorts in order for the examiner to adequately visualize the entire lower extremity including the thigh, knee, lower leg, ankle and foot. When observing a patient’s gait, the following events of the gait cycle related to dynamic foot function should be documented: • Foot position at initial contact – heel strike, flatfoot, or toe strike • Foot alignment in midstance – varus/valgus in the coronal plane and internal/ external in the transverse plane (foot progression angle) • Foot alignment at toe-off – varus/valgus in the coronal plane and dorsi-/plantarflexed in the sagittal plane • Foot clearance in swing phase
Radiographic Evaluation Radiographic images for evaluation of foot and ankle deformities must be routinely performed with the child fully weight bearing. Non-weight-bearing images are not of value in assessing foot segmental alignment and should not be utilized. Three standardized views should always be obtained: • AP foot • Lateral foot • AP ankle
Foot and Ankle Motion in Cerebral Palsy
1135
Table 1 Radiographic measurements of normative values for feet in children Radiographic measurements: quantitative and categorical definitions Normal (mean Abnormal high value 3.75) (> mean +1 SD) Hindfoot Tibiotalar angle (degrees) 1.1 3.75 Eversion Calcaneal pitch (degrees) 17 6.0 Calcaneus Tibio-calcaneal angle 69 8.4 Equinus (degrees) Talocalcaneal angle (degrees) 49 6.9 Eversion Midfoot Naviculo-cuboid overlap (%) 47 13.8 Pronation Talonavicular coverage angle 20 9.8 Abduction (degrees) Lateral talo-first metatarsal 13 7.5 Pronation angle (degrees) Forefoot Anteroposterior talo-first 10 7.9 Abduction metatarsal angle (degrees) Metatarsal stacking angle 82 9 Supination (degrees) Columns Medial-lateral column ratio 0.9 0.3 Abduction
Abnormal low value (> mean +1 SD) Inversion Equinus Calcaneus Inversion Supination Adduction Supination
Adduction Pronation
Adduction
When analyzing radiographs and designating deformities, it is best to divide the foot into three segments and two columns as was described earlier in this chapter, then determining the relative alignment of each segment and the relative length of each column. A comprehensive technique of quantitative segmental analysis of the foot and ankle has been developed by Davids et al. (2005) and was derived from the foot model originally developed by Inman and colleagues (Inman et al. 1981). The technique of quantitative segmental analysis was developed from normative values of a cohort of 60 normal feet in children between ages of 5 and 17 years old, utilizing ten radiographic measurements to determine the alignment of the segments and the lengths of the columns of the foot and ankle (Davids et al. 2005; Table 1).
Quantitative Gait Analysis The calculation of foot and ankle kinematics and kinetics involves modeling assumptions and approximations concerning the relationship between the skin markers and the underlying skeletal anatomy. The standard ankle and foot model most commonly used in clinical gait analysis was developed in the 1980s, uses
1136
J.R. Davids and S.A. Tabaie
markers at the malleoli and forefoot, and considers the foot as a single segment (Perry 1992). It is assumed that the foot segment is rigid from the hindfoot to the forefoot. Ankle motion in the sagittal plane is calculated from the location of the foot axis relative to the tibial axis. Any movement between the three segments of the foot (i.e., hindfoot to midfoot, midfoot to forefoot) that occurs between the malleolar and forefoot markers is captured by this simple foot model and described as ankle motion. Significant measurement artifact occurs when the normal foot segmental alignment is disrupted (i.e., equinoplanovalgus foot malalignment in children with CP). This artifact creates apparent discrepancies within the diagnostic matrix between the data derived from the physical examination, observational gait analysis, and quantitative gait analysis. Failure to appreciate the causes for these apparent discrepancies may result in confusion for clinicians and compromise clinical decision-making. Technological improvements have allowed for the development of more sophisticated, multisegment foot models that more accurately approximate the complex anatomy and biomechanics of the foot (MacWilliams et al. 2003). However, these models are difficult to apply to children with CP, because of small foot size and deformity. Additionally, the dynamic EMG, which is part of the quantitative gait analysis, is very relevant in the evaluation foot deformities in children with CP (Sutherland 1993). The incorporation and use of surface/fine-wire EMG provides information on the timing of muscle activity during the gait cycle (Hoffer et al. 1985; Scott and Scarborough 2006). This data is imperative in determining: • The relative activity of tibialis anterior and posterior muscles in both stance and swing phases • Selection of a particular muscle-tendon unit for lengthening or transfer The final component of quantitative gait analysis is dynamic pedobarography, which measures the spatial and temporal distribution of force over the plantar aspect of the foot during the stance phase of the gait cycle (Jameson et al. 2008; Davids 2009a). Pedobarography provides the following information regarding the patient’s dynamic foot function: • Foot contact patterns • Foot pressure distribution and magnitude • Progression of the center of pressure through the foot As described above, foot function during gait in children with CP is disrupted by several patterns of skeletal segmental malalignment that can ultimately affect all three rockers in stance phase. This biomechanical disruption has been termed lever arm deficiency and is best characterized by the center of pressure progression (COPP) relative to the foot. Deviations in the location and duration of the COPP relative to the segments of the foot when compared to normative values of COPP in children can be used to describe common abnormal loading patterns (Davids 2009a; Fig. 12).
Foot and Ankle Motion in Cerebral Palsy
1137
Fig. 12 Examples of abnormal foot loading patterns (each figure is of a left, as if looking down from above, toes at the top, lateral to the left, medial to the right): (a) displacement of the center of pressure progression (COPP – solid red line) medially, describing a valgus loading pattern. (b) Displacement of the COPP laterally, resulting in a varus loading pattern. (c) Displacement of COPP distally, describing an equinus loading pattern). (d) Displacement of the COPP proximally (i.e., excessive duration of the COPP in the hindfoot segment, reflected by increased thickness of the red line) resulting in a calcaneus loading pattern
• Valgus loading pattern – Displacement of the COPP medially due to an everted, abducted, or pronated segmental malalignment of the foot segment • Varus loading pattern – Displacement of the COPP laterally, which is usually due to an inverted, adducted, or supinated segmental malalignment of the foot segment • Equinus loading pattern – Prolonged duration of the COPP in the forefoot segment • Calcaneus loading pattern – Prolonged duration of the COPP in the hindfoot segment
Levels of Deformity Foot and ankle deformities in children with CP are sequential and progressive with growth and development. The deformities can be classified into three levels (Davids 2010): • Level I – Dynamic soft tissue imbalance with no skeletal pathoanatomy
1138
J.R. Davids and S.A. Tabaie
• Level II – Fixed or myostatic soft tissue imbalance with flexible and correctable skeletal segmental malalignment • Level III – Fixed myostatic soft tissue imbalance with associated structural skeletal deformities Understanding the levels of deformity in children with CP can guide in both the nonoperative and surgical interventions for these patients.
Treatment Principles Management of foot and ankle deformities in children with CP can consist of both nonsurgical and surgical modalities. Treatment options include (Davids et al. 2004, 2007; Gage 1994, 1995, 2004): • Orthotics – May be used to protect the outcome of a surgical procedure during the healing and rehabilitation phases, to prevent the development or worsening of musculoskeletal deformities with growth, and to improve gait. – When the goal is to improve gait, the physician should clearly identify the gait deviations and functional deficits to be addressed by the orthosis. – Clinicians must be familiar with the biomechanical characteristics and clinical indicators of six common orthotic designs (Table 2): 1. Foot orthosis 2. Supramalleolar orthosis 3. Posterior leaf spring orthosis 4. Articulating ankle-foot orthosis 5. Floor-reaction ankle-foot orthosis 6. Solid ankle-foot orthosis • Pharmacologic/Neurosurgical – Level I deformities resulting from dynamic soft tissue imbalance with no skeletal deformities can be treated with interventions to manage muscle tone and spasticity (Preiss et al. 2003; Boyd et al. 2000; Bjornson et al. 2007). 1. Pharmacologic – botulinum toxin (or phenol) injection 2. Neurosurgical – selective dorsal rhizotomy or intrathecal baclofen – Early management with the abovementioned options is favored to avoid development of fixed deformities of the muscle-tendon unit.
Surgical Treatment Goals Surgical management to correct foot and ankle deformities in children with CP may be selected to improve gait or function, shoe wear, and cosmesis. These goals may be
Dorsiflexion to neutral
Dorsiflexion to neutral thigh-foot angle 15 extension
Normal
Normal
Normal
Extension 20
Normal
Normal
Normal
Extension 30
Articulating AFO
Solid AFO
FRAFO
Moderate, partially correctible
Moderate, partially correctible
Mild, correctible
Foot Mild, correctible Mild, correctible Mild, correctible
Absent heel strike
Absent heel strike Absent heel strike Absent heel strike
Normal
Increased plantar flexion, increased knee extension (mild) Increased plantar flexion, increased knee extension, or increased knee flexion (mild) Increased dorsiflexion, increased knee flexion, increased hip flexion
Normal
Normal
Gait deviation Initial contact Midstance Normal Normal
Increased dorsiflexion
Increased plantar flexion
Normal
Normal
Normal
Terminal stance Normal
Increased dorsiflexion
Increased plantar flexion Increased plantar flexion Increased plantar flexion
Normal
Swing phase Normal
UCBL University of California at Berkeley Laboratory, PLSO posterior leaf spring orthotic, AFO ankle-foot orthotic, FRAFO floor reaction ankle-foot orthotic
Dorsiflexion to 5
Dorsiflexion to 5
Normal
Normal
Normal
Supramalleolar orthosis PLSO
Ankle Normal
Knee Normal
Hip Normal
Orthosis UCBL
Indications for use of orthoses to improve gait in children with cerebral palsy Physical exam
Table 2 Indications for the use of common orthoses to improve gait in children with CP
Foot and Ankle Motion in Cerebral Palsy 1139
1140
J.R. Davids and S.A. Tabaie
achieved by surgical procedures that are designed to improve foot shape (Davids 2010; Sutherland 1993; Mosca 1998). • Improved foot shape following soft tissue and skeletal surgery can restore the stability function of the foot during: – The second rocker in midstance and the skeletal lever arm function of the foot during the third rocker in terminal stance • Cosmetic improvements following foot surgery are related to: – Improved visual assessment of static standing foot alignment – restoration of the medial longitudinal arch and toe alignment – Improved foot progression angle during stance phase Additionally, it is presumed that surgically improved foot shape can correct pain by improving foot loading and stability in stance phase. In regard to pain, it is important to make the distinction between the younger children with CP who may tolerate mild or moderate foot deformities versus the older teenage/adult population. Deformities that may be tolerated in a younger CP population are often poorly tolerated in teenage and adult life as the body weight increases, leading to a greater magnitude of the abnormal loading. The cumulative effect of the magnitude of abnormal loading results in premature degenerative changes of the joints of the foot and ankle. Presumptively, surgery to improve foot shape in childhood will improve the loading of the foot and decrease the possibility of early degenerative arthritis in adulthood.
Surgical Treatment Techniques As noted earlier in this chapter, there are three common coupled foot and ankle segmental malalignment patterns in children with spastic-type CP: equinus, equinoplanovalgus, and equinocavovarus. Additionally, hallux valgus is a common secondary malalignment that may be associated with any of the three principle malalignment patterns. The surgical options for these coupled segmental deformities are based on the level of deformity and will be discussed below. • Equinus – Pure plantar flexor malalignment is usually the consequence of overactivity (level I) or tightness (level II) of the ankle plantar flexor muscle group. Assessment of foot segmental alignment with plain radiographs is essential to establishing that there are no associated deformities at the level of the midor forefoot (Davids 2010). – Surgical management starts with level II deformities. – Careful assessment is required to determine the relative contributions of the gastrocnemius and soleus muscles to the fixed shortening of the plantar flexor group. 1. This is best achieved by assessing ankle dorsiflexion range of motion with the knee both flexed and extended.
Foot and Ankle Motion in Cerebral Palsy
1141
2. Isolated limitation with the knee flexed suggests involvement of the soleus muscle, limitation with the knee extended suggests involvement of the gastrocnemius muscle, and limitation regardless of knee position represents involvement of both muscles. – The goal of surgical lengthening of the plantar flexor group is to achieve five degrees of dorsiflexion when the knee is extended. 1. When only the gastrocnemius is involved, a selective fractional lengthening is best performed proximally at the level of the muscle belly. 2. When both muscles are involved and 15 or less of correction is required, selective fractional lengthening mid-calf at the level of the myotendinous junction is preferred. 3. When both muscles are involved and greater than 15 of correction is required, nonselective lengthening distally at the Achilles tendon level is necessary. – In most cases, a fractional lengthening at the level of the muscle belly or myotendinous junction is sufficient. However, with proper patient selection and careful surgical technique, all three techniques described above may be effective and excessive lengthening and weakness of the ankle plantar flexor muscle groups can be avoided (Etnyre et al. 1993; Tylkowski et al. 2009). • Equinoplanovalgus – This malalignment pattern, which is often referred to as flatfoot, is the consequence of overactivity (level I) or tightness (level II) of the ankle plantar flexor and evertor muscle groups. Physical examination is needed to determine whether the muscle-tendon unit deformities are dynamic or mysostatic. Fixed skeletal segmental malalignment (level III) is assessed with plain radiographs which are essential for preoperative planning (Davids 2010). – As with equinus malalignment, in this pattern orthopedic surgical intervention starts with level II deformity and is usually seen in children between 4 and 7 years of age. 1. The treatment choice is lengthening of the ankle plantar flexor muscle group, as described earlier in the equinus section. 2. Additionally, transfer of the peroneus brevis muscle to the peroneus longus muscle should be performed when fixed deformity of the former is present. 3. At the time of surgery, it is essential to determine that normal foot segmental alignment has been restored following soft tissue surgeries (Fig. 13). – Failure to restore normal skeletal alignment with soft tissue surgery alone reclassifies the deformity to level III (Davids 2009b; Mosca 1996, 1998; Yoo et al. 2005; Danko et al. 2004). 1. The primary procedure used to correct level III equinoplanovalgus malalignment is lateral column lengthening, which can be performed at the following locations in the foot: (a) Neck of the calcaneus (b) Calcaneocuboid (CC) joint (c) Body of the cuboid 2. Lateral column lengthening usually allows for correction of all three segments of the foot, most likely secondary to ligamentotaxis (Fig. 14). (a) The lengthening is approximately 1–2.5 cm. (b) Interposition grafting with tricortical iliac crest or allograft.
1142
J.R. Davids and S.A. Tabaie
Fig. 13 The technique for the intraoperative stress assessment of foot segmental alignment. The assistant stabilizes the knee while the surgeon loads the foot and ankle with the foot pusher. A fluoroscopic image is taken in the loaded position and is used to assess the segmental alignment
(c) Lengthening through the CC joint (arthrodesis) is often done in children over 12 years of age. Internal fixation should be used when the graft size is greater than 1.5 cm. in order to promote early mobilization and minimize late graft collapse during phases of healing. 3. Following lateral column lengthening for a level III equinoplanovalgus foot deformity, assessment of the medial column is performed. (a) If there is residual forefoot varus deformity, or if the medial column is hypermobile in the sagittal plane (i.e., unmasked deformities), then a plantar flexion osteotomy (or a combination of both) is made in the medial cuneiform or base of the great toe metatarsal (if the child is skeletally mature. (b) If there is residual forefoot abduction deformity (i.e., incomplete correction), then a talonavicular arthrodesis needs to be performed. (c) If lengthening of the lateral column fails to correct the hindfoot deformity, then arthrodesis of the subtalar, talonavicular, and calcaneocuboid joints are required to achieve optimal alignment of the foot. Fortunately, this is rarely required, even for feet with significant malalignments. • Equinocavovarus – This malalignment pattern that is often referred to as varus foot is usually the consequence of overactivity (level I) or tightness (level II) of the ankle plantar flexor and invertor muscle groups. As with the other malalignment patterns, physical examination is needed to determine whether the muscle-tendon unit deformities are dynamic or mysostatic. Fixed skeletal
Foot and Ankle Motion in Cerebral Palsy
1143
Fig. 14 Correction of moderate level III equinoplanovalgus deformity. (a, b) Lateral and AP radiographs in a child with CP showing equinoplanovalgus segmental malalignment. (c, d) Lateral and AP radiographs of the foot after gastrocsoleus fractional lengthening and lateral column lengthening through the neck of the calcaneus
segmental malalignment (level III) is assessed with plain radiographs and is essential for preoperative planning (Davids 2010). – In children over the 6 years of age with equinocavovarus foot malalignment, surgical management with muscle-tendon transfer can be an option. 1. Information from the physical exam, kinematics, kinetics, dynamic EMG, and pedobarography are used to determine the relative contributions of the tibialis anterior and tibialis posterior muscles to the dynamic varus deformity that occurs during stance and swing phases (Sutherland 1993; Scott and Scarborough 2006; Hoffer et al. 1985).
1144
J.R. Davids and S.A. Tabaie
(a) Split transfer of the tibialis anterior muscle should only be performed when there is ankle dorsiflexion appreciated in midstance. (b) Split transfer of the tibialis posterior should only occur when the dynamic EMG shows the timing of the activation of this muscle corresponds with the presence of varus malalignment during specific subphases of the gait cycle. (c) When kinematics and dynamic EMG assessments are not available, dynamic varus deformity is best treated by concomitant split transfer of the tibialis anterior muscle and fractional lengthening of the tibialis posterior muscle (Barnes and Herring 1991). – Level II deformity involves sequential correction of the hindfoot and midfoot soft tissue contractures (Fig. 15). 1. Correction of the deformity is performed by fractional lengthening of the following muscle groups: (a) Ankle plantar flexor muscle group
Fig. 15 Technique for the correction of level II equinocavovarus deformity. (a) Three incisions are used to lengthen the appropriate soft tissue structures. (b) The plantar fascia and intrinsic muscles of the foot are released through the plantar incision. (c) Fractional lengthening of the abductor hallucis muscle is performed through the distal medial incision. (d) Fractional lengthening of the gastrocsoleus muscle group is performed through the medial calf incision. (e) Fractional lengthening of the tibialis posterior muscle is performed through the same medial calf incision. The flexor digitorum and hallucis longus muscles can also be lengthened, when necessary, though this incision
Foot and Ankle Motion in Cerebral Palsy
1145
(b) Tibialis posterior muscle (c) Flexor hallucis and digitorum longus muscles – rarely necessary (d) Abductor hallucis (e) Plantar fascia and short intrinsic muscle of the foot 2. At the time of surgery, it is essential to determine the normal foot segmental alignment has been restored following the soft tissue surgeries. (a) Confirmation of normal alignment is best achieved with intraoperative stress radiographs of the foot as was described earlier in the chapter. (b) Failure to restore normal skeletal alignment should result in reclassification of the deformity to level III. – Correction of level III equinocavovarus segmental malalignment differs from correction of equinoplanovalgus malalignment, because there is not one single skeletal procedure that can achieve adequate correction of all three segments of the foot (Fig. 16).
Fig. 16 Correction of level III equinocavovarus deformity. (a, b) Lateral and AP radiographs of the foot in a child with CP showing equinocavovarus segmental malalignment. (c, d) Lateral and AP radiographs of the foot after gastrocsoleus fractional lengthening, posterior tibialis fractional lengthening, radical plantar fascia release, calcaneal slide osteotomy, and medial column osteotomy through the cuneiform
1146
J.R. Davids and S.A. Tabaie
1. Gross alignment and dynamic loading of the foot may be greatly improved by performing sequential osteotomies that create deformities to compensate for segmental malalignments. (a) Residual hindfoot varus malalignment may be corrected by calcaneal slide or laterally based closing wedge osteotomies (Koman et al. 1993). (b) Residual midfoot supination deformity may be corrected by lateral column shortening through the cuboid. (c) Residual forefoot pronation may be corrected by dorsiflexion osteotomy of the medial column. (d) Residual forefoot supination deformity may be corrected by plantar flexion osteotomy of the medial column. 2. Arthrodesis strategy is reserved for the most severe cases of level III equinocavovarus segmental malalignment. (a) Double arthrodesis of the calcaneocuboid and talonavicular joints may be required to correct extreme midfoot cavus deformity. (b) Triple arthrodesis of the subtalar, calcaneocuboid, and talonavicular joints may be required to achieve optimal foot alignment. • Hallux Valgus – This deformity in children with CP may be the consequence of intrinsic and/or extrinsic factors. The gait deviations secondary to these factors tend to medialize the forces across the great to metatarsophalangeal (MTP) joint, resulting in an external abduction or valgus moment (Davids 2010). – Radiographic assessment is essential for determining the elements and magnitude of deformity at the great to MTP joint and the presence of associated foot segmental malalignments. – Correction of the deformity is indicated to address the following issues: 1. Pain at the great toe MTP joint 2. Treat hygiene problems related to toe positioning 3. Facilitate shoe and orthotic wear – For hallux valgus deformities in this patient population, there is little evidence to support lengthening or release to correct or improve foot intrinsic muscle tightness (level II deformity). 1. Poor results for soft tissue balancing procedures are most likely a consequence of the failure to address significant extrinsic causes of hallux valgus deformity in children with CP. – The preferred treatment of level III hallux valgus deformity in children with CP is great toe MTP arthrodesis (Davids et al. 2001; Bishay et al. 2009; Fig. 17). 1. Due to the proximal location of the physis of the proximal phalanx, this procedure should not be performed in children who have more than 2 years of growth remaining. 2. The shape of the foot and the child’s gait pattern determines optimal alignment of the MTP arthrodesis.
Foot and Ankle Motion in Cerebral Palsy
1147
Fig. 17 Correction of hallux valgus deformity (in association with comprehensive foot segmental malalignment correction with lateral column lengthening through the neck of the calcaneus). (a) Standing AP radiograph of the foot in a child with CP. The great toe deformity consists of metatarsus varus and MTP joint valgus. (b) Intraoperative fluoroscopy AP showing great toe MTP arthrodesis. (c) Standing AP radiograph of the foot 1 year after great toe arthrodesis (note the indirect correction of the increased inter-metatarsal angle)
(a) Correction of the coronal plane should align the phalanges of the great toe with the lesser toes. Sagittal plane alignment of the arthrodesis should include between 15 and 20 of dorsiflexion (relative to the floor when the foot is weight bearing) to facilitate the forefoot rocker.
Summary • The normal interaction of the ankle/foot with the floor is best described by the three rockers (heel, ankle, and forefoot). • There are three common ankle/foot segmental malalignment patterns seen in children with CP: equinus, equinoplanovalgus, and equinocavovarus. • Each type of ankle/foot malalignment can be classified into three levels based upon the presence of dynamic muscle imbalance, fixed or myostatic deformity of the muscle-tendon unit, and fixed skeletal deformities. • Clinical decision-making for the management of the ankle/foot in children with CP involves a diagnostic matrix utilizing data from the clinical history, physical examination, radiographic imaging, and quantitative gait analysis.
1148
J.R. Davids and S.A. Tabaie
• Common orthopedic surgical interventions used to address ankle/foot problems in children with CP include muscle-tendon unit lengthening or transfer and skeletal osteotomies or arthrodeses. The goals of these surgeries are to rebalance muscle activity, restore range of motion, and realign the skeletal segments of the foot. Acknowledgments Sean Brown, M.A. for assistance in preparation of figure.
References Barnes MJ, Herring JA (1991) Combined split anterior tibial-tendon transfer and intramuscular lengthening of the posterior tibial tendon. Results in patients who have a varus deformity of the foot due to spastic cerebral palsy. J Bone Joint Surg Am 73(5):734–738 Bishay SN, El-Sherbini MH, Lotfy AA, Abdel-Rahman HM, Iskandar HN, El-Sayed MM (2009) Great toe metatarsophalangeal arthrodesis for hallux valgus deformity in ambulatory adolescents with spastic cerebral palsy. J Child Orthop 3(1):47–52 Bjornson K, Hays R, Graubert C, Price R, Won F, McLaughlin JF, Cohen M (2007) Botulinum toxin for spasticity in children with cerebral palsy: a comprehensive evaluation. Pediatrics 120(1):49–58 Boyd RN, Pliatsios V, Starr R, Wolfe R, Graham HK (2000) Biomechanical transformation of the gastroc-soleus muscle with botulinum toxin A in children with cerebral palsy. Dev Med Child Neurol 42(1):32–41 Danko AM, Allen B Jr, Pugh L, Stasikelis P (2004) Early graft failure in lateral column lengthening. J Pediatr Orthop 24(6):716–720 Davids JR (2009a) Normal function of the ankle and foot: biomechanics and quantitative analysis. In: McCarthy JJ, Drennan JC (eds) Drennan’s the child foot and ankle. Lippincott Williams and Wilkins, Philadelphia, pp 54–63 Davids JR (2009b) Orthopaedic treatment of foot deformities. In: Schwartz M, Gage J, Koop S, Novacheck T (eds) The identification and treatment of gait problems in cerebral palsy. MacKeith Press, London, pp 514–533 Davids JR (2010) The foot and ankle in cerebral palsy. Orthop Clin North Am 41(4):579–593 Davids JR (2016) Biomechanically based clinical decision making in pediatric foot and ankle surgery. In: Sabharwal S (ed) Pediatric lower limb deformities. Springer, Cham, pp 153–162 Davids JR, Mason TA, Danko A, Banks D, Blackhurst D (2001) Surgical management of hallux valgus deformity in children with cerebral palsy. J Pediatr Orthop 21(1):89–94 Davids JR, Ounpuu S, DeLuca PA, Davis RB 3rd (2004) Optimization of walking ability of children with cerebral palsy. Instr Course Lect 53:511–522 Davids JR, Gibson TW, Pugh LI (2005) Quantitative segmental analysis of weight-bearing radiographs of the foot and ankle for children: normal alignment. J Pediatr Orthop 25(6):769–776 Davids JR, Rowan F, Davis RB (2007) Indications for orthoses to improve gait in children with cerebral palsy. J Am Acad Orthop Surg 15(3):178–188 Davis RB, Jameson E, Davids JR, Christopher LM, Rogozinski BM, Anderson JP (2007) The design, development, and initial evaluation of a multisegment foot model for routine clinical gait analysis. In: Smith P, Harris GF, Marks R (eds) Foot and ankle motion analysis: clinical treatment and technology. CRC Press, Boca Raton, pp 425–444 Etnyre B, Chambers CS, Scarborough NH, Cain TE (1993) Preoperative and postoperative assessment of surgical intervention for equinus gait in children with cerebral palsy. J Pediatr Orthop 13(1):24–31 Gage JR (1994) The role of gait analysis in the treatment of cerebral palsy. J Pediatr Orthop 14(6):701–702 Gage JR (1995) The clinical use of kinetics for evaluation of pathologic gait in cerebral palsy. Instr Course Lect 44:507–515
Foot and Ankle Motion in Cerebral Palsy
1149
Gage JR (2004) Orthotics: a comprehensive clinical approach. The treatment of gait problems in cerebral palsy. MacKeith Press, London, pp 273–285 Hoffer MM, Barakat G, Koffman M (1985) 10-year follow-up of split anterior tibial tendon transfer in cerebral palsied patients with spastic equinovarus deformity. J Pediatr Orthop 5(4):432–434 Inman VT (1966) The human foot. Manit Med Rev 46(8):513–515 Inman VT (1969) The influence of the foot-ankle complex on the proximal skeletal structures. Artif Limbs 13(1):59–65 Inman VT, Ralston J, Todd F (1981) Human walking. Williams & Wilkins, Baltimore Jameson EG, Davids JR, Anderson JP, Davis RB 3rd, Blackhurst DW, Christopher LM (2008) Dynamic pedobarography for children: use of the center of pressure progression. J Pediatr Orthop 28(2):254–258 Koman LA, Mooney JF 3rd, Goodman A (1993) Management of valgus hindfoot deformity in pediatric cerebral palsy patients by medial displacement osteotomy. J Pediatr Orthop 13(2):180–183 MacWilliams BA, Cowley M, Nicholson DE (2003) Foot kinematics and kinetics during adolescent gait. Gait Posture 17(3):214–224 Mosca VS (1995) Calcaneal lengthening for valgus deformity of the hindfoot. Results in children who had severe, symptomatic flatfoot and skewfoot. J Bone Joint Surg Am 77(4):500–512 Mosca VS (1996) Flexible flatfoot and skewfoot. Instr Course Lect 45:347–354 Mosca VS (1998) The child’s foot: principles of management. J Pediatr Orthop 18(3):281–282 Perry J (1992) Gait analysis: normal and pathologic function. Slack, Thorofare Ponseti IV, El-Khoury GY, Ippolito E, Weinstein SL (1981) A radiographic study of skeletal deformities in treated clubfeet. Clin Orthop Relat Res (160):30–42 Preiss RA, Condie DN, Rowley DI, Graham HK (2003) The effects of botulinum toxin (BTX-A) on spasticity of the lower limb and on gait in cerebral palsy. J Bone Joint Surg Br 85(7):943–948 Scott AC, Scarborough N (2006) The use of dynamic EMG in predicting the outcome of split posterior tibial tendon transfers in spastic hemiplegia. J Pediatr Orthop 26(6):777–780 Stevens PM (1988) Effect of ankle valgus on radiographic appearance of the hindfoot. J Pediatr Orthop 8(2):184–186 Sutherland DH (1993) Varus foot in cerebral palsy: an overview. Instr Course Lect 42:539–543 Tylkowski CM, Horan M, Oeffinger DJ (2009) Outcomes of gastrocnemius-soleus complex lengthening for isolated equinus contracture in children with cerebral palsy. J Pediatr Orthop 29(7):771–778 Yoo WJ, Chung CY, Choi IH, Cho TJ, Kim DH (2005) Calcaneal lengthening for the planovalgus foot deformity in children with cerebral palsy. J Pediatr Orthop 25(6):781–785
The Arm Pendulum in Gait Jaques Riad
Abstract
Because of the physical length of the arms and the relatively large range of motion in the shoulder and elbow in gait, any deviation from normal is detected immediately and attracts attention. The deviation can consist of increased flexion in the elbow, decreased range of motion, movement out of phase with the lower extremities, and asymmetry between the movement of right and left arms, either in isolation or in many more or less noticeable combinations. Although human evolution means that we no longer walk on our arms, arm movement has impact on our stability, balance, and appearance while walking. In addition, we can carry things, make gestures, or do other things with the arms and hands while walking. Despite an evolution toward corticospinal control of arm and hand movements, quadrupedal limb coordination persists during locomotion. We do not think about how we coordinate our arms and legs when walking. It just happens. Individuals with deformity, limited range of motion, or movement disorders affecting the arm show a disturbance of the normal arm pendulum in gait. It can be difficult to understand the consequences of the primary pathology and the influence on the movement pattern as well as the possible development of compensation mechanisms. Studying the arm pendulum is important for diagnosis and treatment and to follow progression over time. In addition, our sensitivity to deviations from normal highlights the importance of arm movement for communication and appearance.
Keywords
Arm movement • Gait • Motion analysis • Upper extremity • Cerebral palsy • Symmetry J. Riad (*) Skaraborg Hospital Skövde, Skövde, Sweden e-mail: [email protected] # Springer International Publishing AG, part of Springer Nature 2018 B. Müller, S.I. Wolf (eds.), Handbook of Human Motion, https://doi.org/10.1007/978-3-319-14418-4_56
1151
1152
J. Riad
Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arm Pendulum and Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mechanical Effects of Arm Movement on Gait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How Information on Arm Pendulum Can Help in Medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Psychological and Social Aspects of Deviation in Arm Pendulum . . . . . . . . . . . . . . . . . . . . . . . Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cross-References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1152 1153 1155 1155 1156 1160 1163 1163 1163
Introduction Smooth and symmetrical arm movements in gait are expected and part of the general movement pattern in human locomotion. The movement in gait sends signals about the walker’s personality, health, and functional status. Symmetry of movement in particular is important for appearance and well-being. Deviation from normal draws attention to abnormalities and implies poor and inadequate function. Asymmetry of movements in gait is immediately apparent to the beholder and can be or become a physiological and psychological problem for the walker. Arm movements in gait, as well as movements of the lower extremity, may be difficult to define and understand. Is the movement pattern primarily an expression of functioning at the lowest possible energy cost? Or are there secondary changes as a consequence of impaired control, injury, deformity, or any other cause? Even tertiary changes might develop owing to previous treatment or progressive deformity with continued growth in children. Obviously, it is essential to separate and identify these differences to make it possible to plan follow-up and treatment. It goes without saying that it is not ideal to treat the secondary or compensatory changes if the primary cause itself can be addressed. This is especially true when the treatment is irreversible, such as surgical interventions. Arm and hand function are essential for independence, something that is in focus in the modern industrial society and crucial for survival of the family in the developing world. In addition, upper extremity movement and function reflect personality and – along with facial expression – contribute strongly in human communication and appearance (Chouchourelou et al. 2006; Ikeda and Watanabe 2009; Meeren et al. 2005; Montemare 1987; Schneider et al. 2014). However, arm movements during gait also play a role in locomotion, contributing to stability of the trunk and keeping energy costs low. Therefore, arm pendulum in gait is of importance. Braune and Fischer included arm movement in their description “The gait of humans” from 1895, but not much attention has been directed specifically toward arm pendulum (Braune 1895). The purpose and mechanism of arm pendulum are still under discussion, although modern techniques have given us a better understanding and to some extent allowed us to appreciate their medical value.
The Arm Pendulum in Gait
1153
State of the Art Different models for analysis of the three-dimensional (3D) motion of the upper extremity have been developed, although overall the system is largely consistent with lower extremity gait analysis (Kadaba et al. 1990). A 3D analysis provides more detailed information than a two-dimensional assessment (video). Nevertheless, video capturing provides a lot of information on the dynamics and appearance of the movement of a moving subject. Video can be especially useful to provide information about transitions such as gait initiation and accelerating to walking or running at a higher speed. Arm posturing with increased elbow flexion and decreased range of motion in the sagittal plane (flexion/extension) is a typical observation in video captures when a child with hemiplegic cerebral palsy (CP) walks or starts running (Fig. 1). Hemiplegic CP is defined as increased muscle tone (spasticity) in the arm and the leg on either the right or the left side and is caused by a brain injury before the age of 2 years. However, the data from a video are difficult to quantify, and the exact degree of flexion and the range of motion can only be measured with a 3D analysis. In addition, the video assessment naturally does not include the rotational profile (i.e., the third dimension).
Fig. 1 Arm posture. In hemiplegic cerebral palsy, the typical one-sided spastic position is with elbow flexion, forearm pronation, and palmar flexion and with little range of motion when walking
1154
J. Riad
Previously, only the lower extremity – what Perry called the locomotor system – was segmented for calculations (two feet, two shanks, two thighs, and the pelvis) (Perry 1992). The head, arms, and trunk were treated as a single unit, the passenger segment. With higher computer capacity, most human motion analysis laboratories now segment also this passenger segment into the head, the trunk, two upper arms, two forearms, and two hands. For recording purposes, the segments are defined by several retroflective markers, which are attached to the skin at specific anatomical landmarks. Several cameras identify the markers when the person walks in the laboratory, and movements relative to other segments can be calculated in three dimensions. If the subject walks on the force plates situated in the floor, information on the forces acting on the body (ground reaction forces) can be obtained. From the movement and the force data, the moments over separate joints can be calculated. Hence, movements (kinematics) and forces (kinetics) are collected, and the information is presented in graphs with the gait cycles for the right and the left side and usually the mean. The gait cycle consists of stance phase (60% of the gait cycle) and swing phase (40%) (Fig. 2). From these data we can obtain exact degrees of movement at any point during the gait cycle. In addition, temporal-spatial data, information on gait speed and step length, stride length, support time, step width, etc., is captured. To obtain a measure of degree of deviation from normal in the upper extremity in gait, the Arm Posture Score (Riad et al. 2011) and the Arm Profile Score (Jaspers et al. 2011) were developed, analogous to the Gait Profile Score (Baker et al. 2009). The deviation is calculated as the root mean square during the entire gait cycle compared to the laboratory’s age- and gender-controlled references (Fig. 3). This score can be useful as a comprehensive assessment of the degree of deviation and to calculate symmetry between both upper and lower extremities. However, the degree of deviation does not tell us what sorts of changes are occurring, merely the degree of deviation. For example, it does not reveal whether the movement is increased or decreased or if the range of motion is increased or decreased relative to normal. It is especially in children with cerebral palsy that modern gait analysis has been evolved as a medical assessment tool. Bonnefoy-Mazure and co-workers, in
Fig. 2 Elbow motion (y-axis) during the gait cycle (x-axis) with the first 60% being the stance phase and the last 40% the swing phase. The gray band in the background is the normal/reference group here; the left (red) and right (blue) sides are plotted with all the gait trials obtained from each side. To the right the means from both sides are illustrated. It is clear that the left side (red) has increased flexion and decreased range of motion
The Arm Pendulum in Gait
1155
Control Patient RMSD
Non-involved side
Involved side Flexion 60.0
60.0
40.0
40.0
20.0
20.0
0.0
0.0
60.0
60.0
40.0
40.0
20.0
20.0
0.0
0.0
GPS Knee Extension Flexion
APS Elbow Extension
Fig. 3 The mean of the sagittal movements of knee and elbow on the involved (hemiplegic) side and the noninvolved side in a child with hemiplegic cerebral palsy. The area between the mean represents the root mean square and is the measurement of deviation
addition, reported comprehensively on upper limb patterns during gait in individuals with CP (Bonnefoy-Mazure et al. 2014). In this chapter we discuss only free arm movements unless stated otherwise. That is, we assume an independent gait without assistive devices other than possible bracing and orthosis of the foot and ankle.
Arm Pendulum and Gait In this chapter, arm pendulum in gait, we will discuss the possible purpose of arm pendulum and the mechanical effects on gait. We will discuss how arm pendulum can be useful in medicine to make diagnoses, how to follow development over time, and how arm pendulum assessment can help in treatment. The medical part will be focused on individuals with neurological impairments, with special emphasis on cerebral palsy, past stroke, and Parkinson’s disease. We will also discuss psychological and social aspects of deviations from normal arm movement in gait.
Mechanical Effects of Arm Movement on Gait Arm pendulum has been reported to serve the purposes of decreasing energy consumption, increasing gait stability, and improving balance (Kuhtz-Buschbeck and Jing 2012; Meyns et al. 2013; Ortega et al. 2008). Some state that arm pendulum is “a relic from quadrupedal walking” without purpose (Jackson et al. 1983; Murray et al. 1967). There is also a debate about the mechanism of arm pendulum.
1156
J. Riad
Muscle activity of the upper limb in gait has been assessed by electromyography (EMG) (Ballesteros et al. 1965; Kuhtz-Buschbeck and Jing 2012) under different conditions: free arms, arms held by the side, bound to the body, and in gait at different speeds. The purpose was to determine whether arm pendulum is active or passive. Muscle activity was noted in normal gait but also in some of the other conditions, and Kuhtz-Buschbeck and Jing concluded that normal arm swing most likely has both active and passive components. In a simulation study with the aim to quantify the components, a significant decrease of arm swing amplitude was found when there was no EMG activity (Goudriaan et al. 2014). The same authors could also show alteration in the pattern (interlimb coordination) when muscle activity was absent. It is not fully clear, but it appears that arm pendulum contributes toward lowering energy consumption, even though there is an energy cost for the swinging (Collins et al. 2009; Elftman 1939). The mechanism is the decrease of vertical ground reaction moment, and the net effect is a reduction of energy consumption by 8% compared to walking with arms constrained (Umberger 2008). Although there are some contradictory results regarding arm swing and lateralmedial stability (Collins et al. 2009), it seems that arm swing contributes to stability (Ortega and Farley 2015; Ortega et al. 2008). Obviously, the small child, the toddler, has the high guard position for stability and balance, and the elderly individual achieves increased trunk stability through increased arm swing (Nakakubo et al. 2014; Kubo and Ulrich 2006). Elderly also had longer recovery time after perturbation to return to steady state gait, with normalized arm swing (Nakakubo 2014). In running, step width is narrower than in walking, and arms swing to minimize energy cost and improve lateral-medial balance (Arellano and Kram 2011). In the neural control of arm pendulum in gait, it appears that a remnant from our quadrupedal days is still in charge of the coordination between the forelimbs and hind limbs, which is mediated by long neurons from the cervical and lumbar spine (Dietz 2011). An uncoupling occurs when the human wants to perform a voluntary arm and hand task, at which point direct control from the cortex takes over. Dietz points out that in patients with Parkinson’s disease, the so-called quadrupedal limb coordination is intact; therefore, verbal, visual, and other stimuli of arm swing can be beneficial in the treatment. In patients who have had a stroke, however, the afferent nerve signal does not lead to an appropriate response (Dietz 2011).
How Information on Arm Pendulum Can Help in Medicine Diagnosis Posturing with the arm – the “high guard” position – is common in toddlers for stabilization (Kubo and Ulrich 2006) and may persist in children with cerebral palsy (CP) (Meyns et al. 2012a). Typically in diplegic CP gait, there is high variability in arm movement and increased arm abduction, as a sign of balance disturbance owing to spasticity and decreased motor control in the lower extremity (Meyns et al. 2012a; Romkes et al. 2007).
The Arm Pendulum in Gait
1157
When attempting to differentiate between hemi- and diplegic cerebral palsy in a young child, the arm posturing often reveals the hemiplegic involvement even in mild cases of CP. In hemiplegic CP, the one-sided spastic position with elbow flexion, forearm pronation, and palmar flexion and with little range of motion (Riad et al. 2007; Romkes et al. 2007) is different from the bilateral elbow flexion pattern in diplegic CP (Fig. 1). It has been found that in children born at term with hemiplegic CP that the arm is more involved than the leg, whereas in children born prematurely, the arm and leg are more equally involved (Uvebrant 2000). Thus, the patient’s history of birth and the clinical sign of arm posturing when walking (and especially when running, which enhances the arm posturing) can contribute toward making a diagnosis. Spastic paraparesis and diplegic CP may be difficult to differentiate. According to Bonnefoy-Mazure and co-workers (Bonnefoy-Mazure et al. 2013), 3D analysis of the upper extremity can help in the diagnoses; they found differences by studying the trunk and arm movement patterns that compensate for gait deviation in the lower extremity. Apart from the fact that patients with spastic paraparesis were older (mean 16.7 years) than the diplegic CP group (mean 12.3 years) and walked slower (mean 1.13 m/s vs. mean 1.42 m/s in the diplegic CP group), no differences between the groups were found: their lower extremity movement in the sagittal plane during gait was indistinguishable in 3D analysis. However, the diplegic group showed increased flexion and more variable arm movements in the arms, similar to the high guard position described in toddlers. In those with spastic paraparesis, compensation for the lower extremity deviation occurred in the trunk and pelvis, which was interpreted as an attempt to stabilize the head. Hence, upper extremity 3D analysis can be helpful in this clinical situation. The gait of patients with Parkinson’s disease is characterized by low velocity, short, shuffling steps and the highly typical reduced arm swing, beside the freezing and difficulties initiating gait. Early diagnosis can be tricky, but is important, since early treatment may have a positive impact on the progression of the disease (Fahn et al. 2004). Reduced arm swing is an early sign of Parkinson’s disease (Lewek et al. 2010), and a study by Mirelman and co-workers reported that use of body-affixed sensors to examine arm swing might provide a new prodromal marker for the disease (Mirelman et al. 2016).
How One Limb Affects the Other Limbs Interlimb coordination is studied to determine if the arms and legs are moving in phase, i.e., if they are normally synchronized, in gait. Meyns and co-workers reported that several children with hemiplegic CP had a 2:1 arm-to-leg ratio on the involved side and a 1:1 ratio on the noninvolved side (1:1 ratio being normal). In children with diplegic CP, also with altered interlimb coordination, increased walking speed improved the synchronization, and the authors suggest that exercise increasing walking speed could be beneficial for the general gait pattern (Meyns et al. 2012b). However, this has neither been tested nor validated, and of course the question arises, what influences what?
1158
J. Riad
In early Parkinson’s disease, newly diagnosed patients show changes in interlimb coordination, and thus coordination analyses help in the diagnosis (Winogrodzka et al. 2005). In addition, it has been suggested that improving interlimb coordination could improve gait in Parkinson’s disease (Dietz 2011). Another interesting study, investigating the effect of elbow contracture on gait revealed significant decreases in gait speed with three different degrees of elbow flexion contractures simulated with a brace in healthy participants (Trehan et al. 2015). In other words, the arm pendulum affects gait also in subjects with no neurological impairment. The arm posture in hemiplegic CP may have an impact on gait and vice versa, and studying the possible influence of different limbs can be quite complex (Lundh et al. 2014). How does the gait respond to the increased elbow flexion and the decreased range of elbow motion? And how does the noninvolved lower extremity/leg respond or rather compensate for the hemiplegic leg? What does this do to the noninvolved arm? The interactions and deviations in relation to each of the two upper and two lower extremities in gait in hemiplegic CP were studied by Lundh (Lundh et al. 2014). By calculating the degree of deviation and symmetry between involved and noninvolved side in 47 adolescents and young adults with hemiplegic CP, four distinct different groups were identified: close to normal, deviations mainly in the leg, deviations mainly in the arm, and deviation in both the leg and the arm. This information can help in diagnoses, as previously described, but can also be useful in long-term follow-up. It is important to identify and understand changes of movement pattern caused by the development of deformity and/or new compensation mechanisms, so as to differentiate between what to address and attempt to treat and what to leave alone.
Natural Development and Treatment of Arm Posture in Children and Young Adults with Cerebral Palsy In children with hemiplegic CP, the arm posture of the involved side is mainly dependent on spasticity and decreased motor control. Children with diplegic CP, however, use their arms more for balance to compensate for lower limb spasticity and impaired motor control. The natural development of arm posturing, elbow flexion in gait, in patients with hemiplegic CP, is that it spontaneously resolves with time (Riad et al. 2007). In a study with 175 individuals (mean age 9.2 years, range 4 to 21 years), there was a significant decrease of elbow flexion with increased age in gait on the hemiplegic side, but not on the noninvolved side. Elbow extension normalized on both sides with increased age. The range of motion increased significantly on the noninvolved side with age, but not on the hemiplegic side (Fig. 4). It was also noted that the variability of elbow flexion in gait decreased with increased age. In the same study, the outcome of elbow flexor lengthening on the natural history of elbow flexion and motion during gait was investigated. Among the 175 children who had severe elbow flexion, some had surgery with elbow flexor lengthening and some did not. The outcome for the two groups was examined by comparison of two
The Arm Pendulum in Gait
1159
Fig. 4 Elbow flexion in gait in 175 individuals with hemiplegic cerebral palsy, with no surgical treatment. The dark area represents the hemiplegic side and reveals decreasing elbow flexion with increasing age, which is the natural development
gait analyses of the same patient over time. At a follow-up time of 3.3 years, the outcome in the surgical and nonsurgical groups was similar, suggesting that elbow flexor surgery had no effect on elbow motion during gait. The explanation for this might be the indication and in particular the individuals and their treatment goals. In addition, children with hemiplegic CP born at term has a higher degree of brain abnormalities, which can have implications when considering surgical correction of arm posture and hand and arm function, since these children would not be expected to cooperate in the rehabilitation program as compliantly as those with normal mental development. Taken together, these observations highlight how important it is to inform the child and concerned parents about the natural history of elbow flexion, and that one should be careful with early surgical intervention. The indications for and optimal timing of surgical intervention are not clear. Attention should be directed toward the child’s possible concerns, and treatment should be offered if the spontaneous correction is not satisfactory (Riad et al. 2007). Treatment with botulinum toxin has been reported to benefit young, mainly highfunctioning adolescents with hemiplegic CP; the patients experienced a very positive cosmetic effect, even though parents and caregivers could not detect any change in movement pattern (Corry et al. 1997). The explanation could be decreased tone, leading to a perception of better control. It is possible that having better control – especially emotional in situations or in transitions of movement (gait initiation, running, etc.) – was beneficial for these patients.
1160
J. Riad
In Treatment of Parkinson’s Disease Verbal instructions including “walking while deliberately swinging the arms” were found to normalize gait pattern regarding gait speed and step length, in addition to arm swing (Behrman et al. 1998). The authors stated that this cognitive strategy could be superior to use of visual cues such as lines drawn in the floor and also compared to hearing cues that are not accessible in daily life. Of course numerous studies have evaluated medical/drug treatment for normalization of arm and gait pattern, and many included symmetry assessments. Here, the 3D analysis can prove to be a useful tool for early diagnosis, evaluation of treatment, and following the progress of the disease. In Treatment of Adults Poststroke In rehabilitation after stroke, handheld training devices for arm function have been reported to have good effect. Virtual reality games have increased patients’ motivation and noticeably facilitated training. After stroke, the arm movement in gait is often abnormal/heavily affected, and this most likely affects exercise and training to a great extent. A study using movable handrails in a treadmill setup demonstrated changes in muscle activation in the lower limb, most likely due to changes in postural stability that occurred when the patient performed arm movements (Stephenson et al. 2010). Interestingly, during swing phase, muscle activity was significantly increased in the tibialis anterior muscle when walking with free/more normal arm movements. This increased muscle activity can help clear the foot in swing phase, which is otherwise a common problem after stroke (Stephenson et al. 2010). Botulinum toxin in elbow flexors resulted in increased gait speed (Esquenazi et al. 2008), and botulinum toxin in the hand, the forearm, and the elbow flexor brachioradialis improved the range of motion in the knee and ankle of “slow striding” patients (Hirsch et al. 2005).
Psychological and Social Aspects of Deviation in Arm Pendulum Our appearance – how we look and move – is of importance and contributes to our behavior, to how we are perceived, and, consequently, to how we feel. One method to study how humans perceive other humans’ movements is to use point-light biological motion sequences. This involves attaching lights to major joints and filming a body in motion. Although the resulting visual sequence shows only the movement patterns of the joints, without showing limbs or body, viewers are readily able to recognize the point lights as representing a human walking. It has been found that the visual detection of motion occurs in the superior temporal sulcus, an area of the brain that is closely connected with the emotional processing area, the amygdala (Johansson 1973). There is evidence that gait characteristics, including arm swing, differentiate emotions (Montemare 1987). There is a clear visual sensitivity to “angry walkers” (Chouchourelou et al. 2006; Ikeda and Watanabe 2009). In
The Arm Pendulum in Gait
1161
addition, arm movement in gait is important for the recognition of individuals, and “exaggerated” movements of the arms facilitated the task of recognition (Hill and Pollick 2000). In their paper “Show me how you walk and I tell you how you feel. . .,” Schneider and co-workers suggested that full understanding of emotion perception requires the investigation of dynamic representation and means of expression other than the face (Schneider et al. 2014). It has been reported that self-esteem and self-concept can be affected by physical impairment even in highly functioning patients with unilateral cerebral palsy (Russo et al. 2008). Arm posturing and gait deviations may be perceived as cosmetic and social impediments when the individual enters adolescence and becomes more selfconscious (Decety and Grezes 2006; Wake et al. 2003). In a group of high-functioning adolescents and young adults with hemiplegic CP, self-esteem was significantly lower than in a control group (Riad et al. 2013). In addition, movement deviations in the upper extremity in gait, mainly elbow flexion, correlate with both self-esteem and sense of coherence. The higher the deviation, the lower the self-esteem and sense of coherence (Fig. 5). Interestingly, no such correlations were seen with deviations in the lower extremity. The conclusion was that movement pattern in the upper extremity should not be assessed solely from a functional perspective. Possible concerns about appearance and influence on selfesteem should also be considered.
Fig. 5 Self-esteem assessed by questionnaire “I think I am” plotted with the Gait Profile Score (GPS) and the Arm Posture Score (APS) expressing degree of deviation from normal. Correlation coefficient ( 0.397) with APS. The higher the deviation, the lower the self-esteem
1162
J. Riad
Symmetry In ancient times, both the Egyptians and the Greeks noted that facial symmetry was important in human judgment of beauty; this has also proven to be true for body symmetry (Brown et al. 2008) (Figs. 6 and 7). In one study, dances performed by symmetrical men, as opposed to asymmetrical men, were rated as more attractive, suggesting that dynamic movement can signal underlying quality independently of static appearance (Brown et al. 2005). It is not well studied or clear what possible impact asymmetrical gait pattern, including arm swing, could have on the individual’s own perceptions and those of other people. Fig. 6 Facial symmetry
Fig. 7 Body symmetry
The Arm Pendulum in Gait
1163
Future Directions In the future, gait assessments should also include the upper extremity, head, and trunk, since this contributes to a more comprehensive understanding of human movement patterns in locomotion. Without doubt, inclusion of total body movement is important for a better understanding, not only for medical purposes but also for a wider understanding of human communication and behavior.
Cross-References ▶ 3D Dynamic Pose Estimation from Marker-Based Optical Data ▶ Clinical Gait Assessment by Video Observation and 2D Techniques ▶ Diagnostic Gait Analysis Use in the Treatment Protocol for Cerebral Palsy ▶ EMG Activity in Gait: The Influence of Motor Disorders ▶ Gait During Real-World Challenges: Gait Initiation, Gait Termination, Acceleration, Deceleration, Turning, Slopes, and Stairs ▶ Gait scores: Interpretations and Limitations ▶ Interpreting Spatiotemporal Parameters, Symmetry, and Variability in Clinical Gait Analysis ▶ Measures to Determine Dynamic Balance ▶ Motor Patterns Recognition in Parkinson’s Disease ▶ Gait Disorders in Persons After Stroke ▶ Slip and Fall Risk Assessment ▶ Spasticity Effect in Cerebral Palsy Gait ▶ Strength Related Stance Phase Problems in Cerebral Palsy ▶ EMG Activity in Gait: The Influence of Motor Disorders ▶ Swing Phase Problems in Cerebral Palsy ▶ Shoulder Joint Replacement and Upper Extremity Activities of Daily Living ▶ Upper Extremity Models for Clinical Movement Analysis
References Arellano CJ, Kram R (2011) The effects of step width and arm swing on energetic cost and lateral balance during running. J Biomech 44:1291–1295 Baker R, Mcginley JL, Schwartz MH, Beynon S, Rozumalski A, Graham HK, Tirosh O (2009) The gait profile score and movement analysis profile. Gait Posture 30:265–269 Ballesteros ML, Buchthal F, Rosenfalck P (1965) The pattern of muscular activity during the arm swing of natural walking. Acta Physiol Scand 63:296–310 Behrman AL, Teitelbaum P, Cauraugh JH (1998) Verbal instructional sets to normalise the temporal and spatial gait variables in Parkinson’s disease. J Neurol Neurosurg Psychiatry 65:580–582
1164
J. Riad
Bonnefoy-Mazure A, Turcot K, Kaelin A, De Coulon G, Armand S (2013) Full body gait analysis may improve diagnostic discrimination between hereditary spastic paraplegia and spastic diplegia: a preliminary study. Res Dev Disabil 34:495–504 Bonnefoy-Mazure A, Sagawa Y Jr, Lascombes P, De Coulon G, Armand S (2014) A descriptive analysis of the upper limb