«A Thesis Presented to The Academic Faculty by Mark Hongchul Sohn In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in ...»
A COMPUTATIONAL FRAMEWORK TO QUANTIFY
NEUROMECHANICAL CONSTRAINTS IN SELECTING
FUNCTIONAL MUSCLE ACTIVATION PATTERNS
The Academic Faculty
Mark Hongchul Sohn
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in the
Woodruff School of Mechanical Engineering
Georgia Institute of Technology
COPYRIGHT © 2015 BY MARK HONGCHUL SOHN
A COMPUTATIONAL FRAMEWORK TO QUANTIFY
NEUROMECHANICAL CONSTRAINTS IN SELECTING
FUNCTIONAL MUSCLE ACTIVATION PATTERNS
Dr. Lena H. Ting, Advisor Dr. David Hu Wallace H. Coulter Dept. of Biomedical Woodruff School of Mechanical Engineering Engineering Emory University and Georgia Institute of Georgia Institute of Technology Technology Dr. Thomas Burkholder Dr. Jun Ueda School of Applied Physiology Woodruff School of Mechanical Georgia Institute of Technology Engineering Georgia Institute of Technology Dr. Magnus Egerstedt School of Electrical and Computer Engineering Georgia Institute of Technology Date Approved: April, 6, 2015 Soli DEO Gloria
ACKNOWLEDGEMENTSI would like to first thank my advisor, Dr. Lena Ting who has been a patient boss, a generous mentor, an inspiring teacher, an admirable colleague scientist, and a sincere friend to me throughout my graduate years. The word grace in a particular worldview of mine refers to a precious gift endowed to one who does not deserve it or have no rights to be endowed, yet solely by the good will of the endower. Lena has always been, and will be, more than I deserve.
I am also grateful to my thesis committee, Dr. Thomas Burkholder, Dr. Magnus Egerstedt, Dr. David Hu, and Dr. Jun Ueda. Their feedback, criticisms have provided a broader aspect to the problem, and have lit the way. Further, I thank Drs. Richard Nichols, Young-Hui Chang, Randy Trumbower, Francisco Valero-Cuevas, and Eric Perreault for leading me into the discussions that were full of insights.
I could not have been where I am now if it were not for the Neuromechanics lab. I thank Lucas McKay who, as my graduate mentor, has taught me how to think, read, write, talk, listen, and become a scientist. I thank Jeff Bingham who always marvels me with his talented skills, and makes me humbled by his noble character. I thank Stacie Chvatal for the warmth in heart she showed me in my times of despairs, helping me to remember that what I have already is enough. I thank Nate Bunderson, Seyed Safavynia, Julia Choi, Keith van Antwerp for their support. I thank Andrew Sawers, Jessica Allen, Yun Seong Song, Kyle Blum, Kim Lang, Aiden Payne, Danny Smith, and Chris Versteeg for always being there for help. I thank Cole Simpson for letting me be assured, once again, that through mentoring one can help another to live and let others live abundant life.
Dr. Jaemin Shin and (soon to be Dr., and a mother) Namin Jeong, who were there from the beginning when I was down on the ground with bare hands. I can never thank them enough for their love and supports.
I have debt of prayers throughout the life to my parents in Korea, Myong Hwan Sohn and Young Soon Shin. I stand in awe of Him who has fulfilled the promises they have received more than 30 years ago here in Atlanta so faithfully. It will be a burden for me for the rest of my life, as I raise my own children, to reflect how great of a parent they were. I would like to acknowledge my sister Hyojin Sohn and my twin brother David Hongsuk Sohn, they are always in my heart.
Lastly, but not the least, I want to thank Jinhee Kim for being there beside me in any circumstances. Ever since she has walked into my life, every day is a miracle, a testimony. I thank her for bringing into our lives our beloved, Yul Junpyo Sohn, who has given me the reasons to take steps forward, who is a reason for my daily breath.
Studying neural control of movement has always amazed me as I witness the beauty of the design that is all so wondrous. Even though all I could seek for my entire life would still be only a tiny bit of the whole, I shall walk this path humbly in recognition of the Designer who reigns from above. All is Yours, and You are my all.
Figure A.4: Computed upper and lower bounds of feasible ranges of muscle activation as a function of the gait cycle with experimental EMG data superimposed 139
Understanding possible variations in muscle activation patterns and their functional implications to movement and motor control is crucial for rehabilitation. Interand intra-subject variability is often observed in muscle activity measured during performance of the same task in both healthy and impaired individuals. However, the extent to which muscle activation patterns can vary under specific neuromuscular conditions and how they may differ functionally are still not well understood. Current musculoskeletal modeling approaches using optimization techniques cannot adequately address such questions because they focus on identifying a unique optimal muscle activation pattern, though many possible patterns exist that could produce the same movement. Therefore, we need an alternative modeling framework to explore and characterize the range of possible muscle activation patterns for a given task and to explain the functional implications of such variations.
Here I developed a novel computational framework that uses a detailed musculoskeletal model to reveal the latitude the nervous system has in selecting muscle activation patterns for a given task with respect to various biomechanical and neural constraints. I specifically focused on an isometric endpoint force generation task relevant to standing balance behavior in cats using a cat hindlimb model. By identifying the explicit bounds on activation of individual muscles defined by biomechanical constraints of the limb and the task, I demonstrate that there exists a wide range of feasible activation patterns that may be sufficient to account for experimentally observed variability. By investigating the possible biomechanical and neural bases of using the same muscle
generalization of function can affect the selection of muscle activation patterns that is not granted by limb biomechanics nor a single optimality criterion. By characterizing the landscape of the solution space with respect to two functional properties, effort and stability, I also demonstrate a possible trade-off in neural selection of muscle activation patterns for a given task that may explain individual differences. In addition, I discuss work where I have extended the method to the dynamic task of human walking. Finally, I present preliminary work showing how the modeling framework developed in this thesis can be used for understanding impaired motor control by considering altered biomechanical and neural constraints.
Neuromechanical principles underlying functional and impaired movements can be elucidated by understanding the allowed variability in muscle activation patterns and evaluating the functional consequences of such variations. Specifically, we may gain valuable insights to explaining individual differences in movement strategies, motor learning, or compensation to neuromuscular disorders. Furthermore, this framework can be useful in developing novel biologically-inspired control principles for robots and effective patient-specific treatments and rehabilitation strategies.
In control of movement, humans respond to unexpected perturbations in a robust manner and flexibly adapt to novel tasks. Yet, how the nervous system achieves such graceful control is not fully understood. In particular, redundancy at multiple levels of sensorimotor transformation poses a degrees of freedom problem (Bernstein 1967) in that there are many ways in how the same motor task can be achieved. Accordingly, substantial inter- and intra-subject variability is observed in muscle activity measured during a movement in both healthy and impaired individuals (Gottlieb 1998; Horak and Nashner 1986; Winter and Yack 1987). Understanding this motor abundance manifested in such allowed variability is essential to elucidating the neuromechanical control underlying functional and impaired movements.
What determines the differences? That is, what determines which muscle activation pattern is to be used for a given task? The main question that has remained unanswered is the degree to which a muscle activation pattern for a given movement is determined by biomechanical or neural constraints. Biomechanical constraints refer to limiting factors owing to the capability of the musculoskeletal system such as the anatomical arrangement of limb musculature, and the task requirements. Musculoskeletal redundancy may allow for ample room in how multiple muscles can be coordinated to achieve a movement under given biomechanical constraints, especially for sub-maximal tasks. Thus, kinematic or morphological differences may not solely account for differences in muscle activity across individuals or trials (Walter et al. 2014). On the other hand, neural constraints refer to the principles or control strategies by which the nervous system selects particular solutions to be used. Neural redundancy may, as well, allow for many functionally equivalent solutions, which may not be necessarily optimal
conditions may affect the possible choices. For example, compromised systems such as weakened muscles or reduced cortico-spinal input due to neurological pathologies may induce a compensatory mechanisms to achieve a functional motor task.
However, it is difficult to understand or interpret variability in muscle activity.
Questions may arise: What are the functional differences between one muscle activation pattern and another? What are the possible changes that can be made to muscle activation patterns in individuals with impairment for recovering function? In order to address these types of questions, it is crucial that we understand possible variations in muscle activation patterns and its functional implications to movement and motor control.
Current musculoskeletal modeling approaches use optimization techniques that cannot adequately address such questions because they focus on identifying a unique optimal muscle activation pattern among the many possible patterns that could produce the same movement. Accordingly, optimal predictions often deviate from measured muscle activity and cannot account for experimental variability (Buchanan and Shreeve 1996; Herzog and Leonard 1991; Thelen and Anderson 2006). Nevertheless, detailed musculoskeletal models can be useful tools because they provide high-resolution access to internal variables, such as muscle force, and definitive causal relationships allowing isolated control over parameters of interest (Bunderson and Bingham in press; Hicks et al.
2015). We currently lack a quantitative modeling framework to explore the range of possible muscle activation patterns for a given task using musculoskeletal models, and to predict the functional implications of such variations.
To this end, this thesis presents a novel computational framework that uses a detailed musculoskeletal model to reveal the latitude the nervous system has in selecting muscle activation patterns for a given task with respect to various biomechanical and neural constraints.
1.1.1 Challenges in studying neural control of movement Studying neural control of movement is a difficult problem because it involves complex interaction among neuromechanical elements at multiple hierarchical levels (Nishikawa et al. 2007; Ting et al. 2012). At the execution level, muscle activity is an output of the nervous system that incorporates control signals from multiple sources such as reflexive feedback from spinal circuits and volitional commands descending from higher centers. On the other hand, the transformation from muscle activation to actual movement occurs in a highly nonlinear manner where forces exerted by active muscles result in net moments at the joints that subsequently produce joint movement or limb endpoint forces (Ting and Chiel in press). Biomechanical constraints such as interaction torques (Gribble and Ostry 1999; van Antwerp et al. 2007) and state dependency of muscles’ force generation on the dynamics of the body (Gordon et al. 1966; Hill 1953;
Rack and Westbury 1969; Zajac 1989) further complicate the problem. What remains largely unknown is the extent to which either biomechanical or neural constraint determines the selection of a particular solution for a given movement.
In particular, understanding the influence of musculoskeletal redundancy on the selection of muscle activation patterns is difficult. Because the number of muscles exceeds the number of joints to be controlled, any movement can be produced with multiple patterns of muscle activation. Accordingly, muscle patterns used during production of similar biomechanical outputs such as joint kinematics or endpoint force varies across individuals and across trials, as measured in substantial variability in electromyography (EMG) (Horak and Nashner 1986; Torres-Oviedo et al. 2006; TorresOviedo and Ting 2007). However, it is difficult to interpret such variability with respect to the biomechanical latitude that the nervous system has when selecting muscle activation patterns. There have been only a few studies that have quantitatively examined
and Valero-Cuevas 2011; Martelli et al. 2015; Martelli et al. 2013).
In addition, the functional criteria by which the nervous system may select a specific muscle activation pattern among many possible is not readily identifiable.
Variability in neuromotor behaviors results naturally from a large space of equivalent solutions (Klein et al. 2010; Prinz et al. 2004; Raphael et al. 2010). Nevertheless, the nervous system seems to coordinate muscles in a specific manner, exhibiting robust patterns across many functional behaviors (Carlsöö 1972; Elble et al. 1994; Mann et al.