B.J. Fregly 9/2014

RESEARCH INTERESTS OVERVIEW The ultimate goal of my research program is to develop “computational rehabilitation” methods that will significantly improve walking function for individuals affected by neurological (e.g., stroke, Parkinson’s disease) and musculoskeletal (e.g., knee osteoarthritis) movement disorders. “Computational rehabilitation” involves combining 1) an understanding of clinical problems that limit walking function with 2) engineering technologies that can model walking function and 3) computational methods that can predict walking function for various interventions under consideration, ideally leading to highly effective rehabilitation prescriptions that are personalized to the needs of each patient (Fig. 1). At the core of this effort is the emerging field of “computational neuromechanics,” which uses computational technology to integrate neural control principles with engineering mechanics principles to analyze, simulate, and predict human movement. This unique approach to personalized treatment planning could significantly improve rehabilitation of walking dysfunction by providing at least three new clinical capabilities: 1) The ability to identify which patients have the capacity to respond to existing treatment approaches.

This capability would help avoid wasting time, effort, and money on treatments that are likely to be ineffective for a particular patient.

2) The ability to customize existing treatment approaches to maximize their efficacy for any particular patient. This capability would increase the likelihood of a positive functional outcome without developing new treatment methods.

3) The ability to design new treatment approaches that may be generalizable to large patient populations or that may require customization to each patient. This capability would yield novel treatments with greater efficacy than existing treatments.

In all three cases, a patient’s response to rehabilitation would be predicted quantitatively rather than qualitatively using objective rather than subjective treatment planning methods.

Figure 1: Overview of the key elements of my research program and how they interact. Challenging rehabilitation problems motivate the use of enabling engineering technologies to develop clinically useful post-treatment predictions of patient function that inform actual treatment. Computational neuromechanics – using computational technology to combine neural control principles with engineering mechanics principles - is at the core of this research approach.

! Stroke! Parkinson’s disease! Knee osteoarthritis! Below-knee amputation! Cerebral palsy! Spinal cord injury

Challenging RehabilitationProblems

! Walking speed improvements! Walking symmetry improvements! Walking motion changes! Muscle coordination changes! Muscle force changes! Joint contact force changes

Clinically Useful Post-treatmentPredictions

! Muscle synergy analysis! Multibody dynamics! Numerical methods! Optimization! Optimal control! Contact mechanics

Enabling EngineeringTechnologies

ComputationalNeuromechanics

pinal cord injury

! Walking! Walking

Clinically Uy

alysis

eering

ComputationalNeuromechanics

UU

B.J. Fregly 9/2014

2

The remainder of this document provides a Motivation for this research direction, a description of the Approach that I am following to achieve my ultimate goal, and finally a brief discussion of Projects I am currently pursuing to make this goal a reality. MOTIVATION

“Doctors have long known that people differ in susceptibility to disease and response to medicines. But, with little guidance for understanding and adjusting to individual differences, treatments developed have generally been standardized for the many, rather than the few.” – National Academy of Engineering1

“The Brain Research through Advancing Innovative Neurotechnologies (BRAIN) Initiative is a new Presidential project aimed at revolutionizing our understanding of the human brain. By accelerating the development and application of innovative technologies, researchers will be able to produce a new dynamic picture of the brain that, for the first time, shows how individual cells and complex neural circuits interact in both time and space. It is expected that the application of these new tools and technologies will ultimately lead to new ways to treat, cure, and even prevent brain disorders.” – National Institutes of Health2

Clinical conditions affecting the neuromusculoskeletal system (e.g., stroke, cerebral palsy, Parkinson’s disease, osteoarthritis) can significantly limit walking function and are primary causes of disability among U.S. adults3. Such conditions cost the U.S. economy billions of dollars annually in medical care and lost productivity3. Furthermore, as walking ability diminishes, quality of life decreases4 and the risk of death increases5,6, making restoration of walking function a critical issue for public health. Current rehabilitation methods for walking-related disorders rely heavily on subjective clinical judgment and have been largely ineffective at restoring normal walking function. These observations underscore a significant knowledge gap regarding development of effective rehabilitation methods as well as an urgent unmet need for these clinical populations. The convergence of two factors makes this the perfect time to address this critical rehabilitation need. The first is the National Academy of Engineering’s identification of “personalized medicine” as one of the Ten Grand Challenges of the 21st Century1. The main idea behind “personalized medicine” is to replace trial-and-error “one size fits all” treatment approaches with new approaches that use patient-specific measurements and/or models to select between competing treatment options or to customize a treatment option to the unique clinical characteristics of the patient. Rehabilitation problems fit well within this context given that patients often present with unique clinical challenges that may not respond well to generic treatment approaches. The second factor is the recent initiation of the BRAIN Initiative funded by the National Institutes of Health and the National Science Foundation2. The goal of the BRAIN Initiative is to find new ways to treat brain-related disorders through the development and use of cutting-edge technologies. Clinical conditions affecting walking function frequently possess a neurological component, making neurorehabilitation of walking an excellent fit for this initiative. Taken together, these two factors provide strong motivation for using technology to develop personalized neurorehabilitation methods that address walking dysfunction. APPROACH Rehabilitation technology can be used to help patients regain normal function by changing the brain’s control strategy (i.e., restoration) or replace lost function by using external assistive devices (i.e., compensation). While compensation can provide mobility, restoration is the desired outcome as it provides the greatest health and quality of life benefits. The question, then, is how technology can be used to facilitate the development of personalized neurorehabilitation treatments that restore walking function to the fullest extent possible. I am using computational neuromechanics as the core technology for developing personalized neurorehabilitation treatments. Computational neuromechanics uses computational technology to combine neural control principles for how the brain controls movement with engineering mechanics principles (specifically multibody dynamics) for how muscle and external forces produce movement. Because of muscle control redundancy (i.e., many more muscle actuators than degrees of freedom in the

B.J. Fregly 9/2014

3

skeleton), engineering mechanics alone does not provide enough constraints to explain how muscles coordinate movement. However, neural control principles can provide additional constraints on the muscle coordination patterns produced by the brain. Furthermore, these additional constraints are patient-specific and can be quantified using a computational method called “muscle synergy analysis”7,8. Muscle synergy analysis decomposes a large number of experimentally measured muscle excitation signals into a much smaller number of independent control signals that can be linearly combined to reconstruct the original muscle excitations. While four or five independent control signals are typically required to reconstruct leg muscle excitation signals during walking9, individuals who have had a stroke or have Parkinson’s disease typically require only two or three10,11, indicating reduced neural control complexity and limitations on achievable muscle coordination patterns.

Figure 2: Overview of my lab’s neuromechanical modeling approach for predicting best achievable walking motions. Top section: A muscle synergy controller varies a small number of experimentally determined independent neural commands (blue curves) to control a larger number of muscle excitations (red curves). Middle section: EMG-to-moment models convert predicted muscle excitations (EMG signals) into net joint moments (green curves). Bottom section: A multibody dynamic walking model whose joints (orange circles) are controlled by predicted net joint moments is used within an optimal control framework to find an achievable low-dimensional set of neural commands that produce the best walking motion possible.

EMG

5

% Gait Cycle

EMG

1

% Gait Cycle

EMG

3

% Gait Cycle

EMG

2

% Gait Cycle

EMG

4

% Gait Cycle

Com

man

d 1

% Gait Cycle

Com

man

d 3

% Gait Cycle

Com

man

d 2

% Gait Cycle

Mom

ent 2

% Gait Cycle

Mom

ent 1

% Gait Cycle

Mom

ent 3

% Gait Cycle

s21 s22 s23 s24 s25 s31 s32 s33 s34 s35s11 s12 s13 s14 s15

Synergy1 Synergy2 Synergy3

Mus

cle

Syne

rgy

Con

trolle

r

c11

Contribution1c21 c22

Contribution2c32

Contribution3c53

Contribution5c42 c43

Contribution4

EMG

-to-M

omen

t Mod

elM

ultib

ody

Wal

king

Mod

el

B.J. Fregly 9/2014

4

By using patient-specific muscle synergy information to control patient-specific multibody dynamic walking models, my lab is seeking to develop neuromechanical walking optimizations that yield unique muscle force estimates and predict the best walking pattern that a patient is capable of achieving (Fig. 2). These optimizations will provide a bridge between a patient’s neural control capabilities and limitations and the patient’s resulting walking function. Optimization predictions will be used to identify the best existing rehabilitation approach and how to implement it or to design new rehabilitation approaches. My lab has already used computational methods to design a new rehabilitation approach - a modified walking motion to offload the diseased medial side of the knee for individuals with medial knee osteoarthritis12. We have also proven that this rehabilitation approach works using experimental data collected from a subject implanted with a force-measuring knee replacement13. To date, this modified walking motion is the only rehabilitation procedure designed by a computational walking model rather than by a clinician. My lab’s neuromechanical optimization approach is consistent with clinical evidence that the brain can be “rewired” (often called “neuroplasticity”) to restore a normal walking motion without restoring normal muscle coordination14,15, a concept called “motor equivalence.” Motor equivalence, which is a logical consequence of muscle force redundancy, suggests that neurorehabilitation should not seek to restore normal muscle coordination but rather to identify the closest achievable coordination strategy that will produce a normal or near normal walking motion. This goal is advantageous since it defines an unambiguous optimization cost function – minimize neural control changes away from the patient’s pre-treatment state – for predicting a patient’s best achievable walking pattern. Exactly how neural control changes should be minimized for each patient is a topic of ongoing research in my lab. PROJECTS My lab is pursuing a number of current projects – both funded and unfunded – to realize my vision of using computational neuromechanics to make neurorehabilitation of walking highly effective and efficient on an individual patient basis. Below is a brief overview of my current research efforts. Two NSF projects (both funded on the first submission) are currently providing the theoretical foundational for developing predictive patient-specific walking optimizations. The first project, entitled “Computational Neuromechanics for Stroke Rehabilitation” ($330,000), is investigating how to implement patient-specific muscle synergy controllers that account for the neural control capabilities and limitations of individuals post-stroke. The project will couple these controllers to a patient-specific dynamic walking model using a novel EMG-to-moment modeling method that eliminates the need for an explicit geometric model of each patient’s leg muscles and bones16. The ultimate goal is to predict the best walking pattern that any individual post-stroke is theoretically capable of achieving. The second project, entitled “CDS&E: A Next-Generation Computation Framework for Predicting Optimal Walking Motion” ($500,000), is exploring how optimal control methods can be used to predict muscle forces and novel walking motions more effectively than with traditional movement optimization methods. Inverse dynamic optimization methods do not fully satisfy the relevant dynamics equations and make it difficult to model foot-ground interactions, but they are computationally fast and do not require control systems to prevent unstable walking models from falling over. In contrast, forward dynamic optimization methods fully satisfy the relevant dynamics equations and make it easier to model foot-ground interactions, but they are computationally slow due to numerical integration and stiffness issues and require control systems to maintain the body upright. Properly formulated optimal control problems may provide the benefits of both approaches simultaneously and allow prediction of post-treatment walking patterns in a highly reliable and computationally efficient manner17. An NIH R01 project entitled “Evaluation of In Vivo Knee Load Predictions using Instrumented Knee Implants” ($2.27 million) is currently providing experimental data for evaluating the reliability of leg muscle and knee joint contact forces estimated using musculoskeletal computer models. This project is the largest NIH R01 grant awarded to a faculty member in the University of Florida College of Engineering in the past 10 years, and it is an international effort involving collaborators from Scripps Clinic in La Jolla, Stanford University, the University of Auckland in New Zealand, Griffith University in Australia, and the University of Melbourne in Australia. Along with model personalization, model validation is one of the primary obstacles to future clinical utilization of neuromusculoskletal computer models18. This project is using movement data collected from subjects implanted for force-measuring knee replacements to validate musculoskeletal model predictions of internal forces in the body19. We are using these unique data sets to evaluate how neural constraints derived from subject-specific muscle synergy information

B.J. Fregly 9/2014

5

can improve muscle excitation and knee contact force predictions generated using subject-specific musculoskeletal models. We have already demonstrated that use of subject-specific neural constraints improves model predictions of knee contact forces during walking20. We are currently exploring how this computational approach can improve the prediction of muscle excitation patterns as well. To encourage greater research effort on musculoskeletal model validation, I have also initiated an annual “Grand Challenge Competition to Predict In Vivo Knee Loads” at the ASME Summer Bioengineering Conference19. The competition follows a “crowdsourcing model” by eliciting voluntary participation by the best researchers in the field from around the world. Each year for the past five years, we have used Stanford’s Simtk.org website (https://simtk.org/home/kneeloads, Fig. 3) to release a new data set collected from a subject implanted with an instrumented knee replacement. Competitors predict knee contact forces during walking in a blinded fashion using their own musculoskeletal modeling methods and submit their predictions to the conference. We release the instrumented implant measurements only after conference abstracts have been submitted. The best submissions are presented in a special session at the conference, and the winning team receives a $1,000 cash prize. Research teams from 12 countries have participated, with no competing team to date producing highly accurate blinded contact force predictions, indicating that significant room for improvement still exists21. From a broader perspective, these data sets have been used in 54 journal articles, 62 conference papers, 3 Ph.D. dissertations, 3 university courses, and 5 successful NIH grant proposals, excluding work performed by my lab or collaborating labs. NIH has been very pleased with our open approach and the substantial impact it has had on the field already. The project will terminate in spring 2015, and I am currently working with my collaborators to develop a renewal proposal for submission in fall 2015.

My lab is also exploring several unfunded projects that dovetail well with my funded projects and are strategic for future grant proposals. One project involves analyzing whether variability in pre-treatment muscle synergies can be used to identify which patients will respond and which will not respond to a novel neurorehabilitation procotol to improve walking ability post-stroke22. If “responders” and “non-responders” can be identified prior to treatment, then this treatment approach can be prescribed when it is most likely to be beneficial. The analyses may also provide valuable clues for how pre-treatment muscle synergy variability can be used to determine the extent to which a patient can modify his or her neural

Figure 3: Map showing website page hits for the human movement data sets and models being released for my NIH R01 “Knee Grand Challenge” project (https://simtk.org/home/kneeloads). These data sets and models are being used by researchers around the world to evaluate musculoskeletal model predictions of leg muscle forces and knee contact forces.

B.J. Fregly 9/2014

6

control strategy in response to treatment. Another stroke rehabilitation project involves the use of transcranial magnetic stimulation (TMS) to probe the neural circuitry of the corticospinal tract during walking and identify which leg muscle excitations (especially for distal muscles) could potentially be increased by the patient’s brain. A neuromechanical walking model of each patient will be constructed, and muscle excitations in the model will be increased at different locations in the gait cycle for muscles with positive TMS probe results. We will predict whether a small set of potentially achievable muscle excitation increases could produce significant improvement in the patient’s walking speed and symmetry. If so, these excitation increases would become targets for therapy (e.g., by using TMS to retrain the brain to “boost” select muscle excitations at select locations in the gait cycle). I am also exploring a new project involving identification of optimal rehabilitation tasks for individuals with Parkinson’s disease (PD). In some cases, individuals with PD have significantly impaired walking function but can still ride a bicycle normally. This surprising observation suggests that the amount of kinematic constraint inherent in a particular locomotion task may have a profound effect on an individual’s neural control capabilities. Movement and muscle excitation data collected during a host of locomotion tasks with different amounts of kinematic constraint will be analyzed to determine how neural control complexity changes across tasks and whether a particular task exists that would drive a patient’s muscle synergies toward the synergies needed to improve walking function. I am also currently supervising two undergraduate honors thesis projects. One is using musculoskeletal modeling to assess whether recurrent falling by individuals with below knee amputation may be caused by a reduction in knee stiffness resulting from loss of calf musculature, and the other is developing a modified muscle synergy analysis approach to investigate whether sensory feedback may explain muscle synergy changes across different walking speeds23. Pursuit of these projects would not be possible without strong collaborations on both the clinical side and the engineering side. My stroke rehabilitation research is performed in collaboration with Dr. Carolynn Patten of the University of Florida Department of Physical Therapy and the Brain Rehabilitation Research Center of Excellence located in the Malcom Randall VA Medical Center in Gainesville. Dr. Patten has both a Ph.D. degree in neuroscience and a P.T. degree and provides clinical and experimental expertise (e.g., TMS methods) for our collaboration. My work on Parkinson’s disease is performed with Dr. Chris Hass in the University of Florida Department of Health and Human Performance. Dr. Hass has a strong track record in PD research and provides clinical and experimental expertise (e.g., split belt treadmill training methods) as well. My research involving instrumented knee implant data and knee osteoarthritis rehabilitation is performed in collaboration with Dr. Darryl D’Lima of Scripps Clinic in La Jolla. Dr. D’Lima is an orthopedic surgeon who has a Ph.D. degree in bioengineering and has pioneered the development of instrumented knee implants. Key engineering collaborators on my NIH R01 project include Dr. Thor Besier of the University of Auckland in New Zealand, Dr. David Lloyd of Griffith University in Australia, Dr. Scott Delp of Stanford University, and Dr. Marcus Pandy of the University of Melbourne in Australia. Key engineering collaborators at the University of Florida include Dr. Scott Banks in the area of fluoroscopic imaging of human joints during movement and Dr. Anil Rao on the application of optimal control methods to neuromusculoskeletal simulations. REFERENCES 1. National Academy of Engineering, Engineering Better Medicines.

http://www.engineeringchallenges.org/cms/8996/9129.aspx. 2. Notice of Informational Conference Calls for Prospective Applicants to NIH BRAIN Initiative Funding

Opportunity Announcements. http://grants.nih.gov/grants/guide/notice-files/NOT-NS-14-005.html.

3. CDC (2005) Prevalence and most common causes of disability among adults --- United States. Morbidity and Mortality Weekly Report 58, 421–426.

4. Praemer A., Furner S., and Rice D.P. (1999) Musculoskeletal Conditions in the United States. American Academy of Orthopaedic Surgeons, Rosemont, IL.

5. Mutikainen S., Rantanen T., Alén M., Kauppinen M., Karjalainen J., Kaprio J., and Kujala U.M. (2011) Walking ability and all-cause mortality in older women. International Journal of Sports Medicine 32, 216–222.

6. Brown J.C., Harhay M.O., and Harhay M.N. (2014) Walking cadence and mortality among community-dwelling older adults. Journal of General Internal Medicine 29, 1263–1269.

B.J. Fregly 9/2014

7

7. Ivanenko Y.P., Poppele R.E., and Lacquaniti F. (2004) Five basic muscle activation patterns account for muscle activity during human locomotion. The Journal of Physiology 556, 267–282.

8. Ting L.H. and Chvatal S.A. (2010) Decomposing muscle activity in motor tasks: methods and interpretation. In: Motor Control: Theories, Experiments, and Applications. Oxford University Press, New York.

9. Cappellini G., Ivanenko Y.P., Poppele R.E., and Lacquaniti F. (2006) Motor patterns in human walking and running. Journal of Neurophysiology 95, 3426–3437.

10. Clark D.J., Ting L.H., Zajac F.E., Neptune R.R., and Kautz S.A. (2010) Merging of healthy motor modules predicts reduced locomotor performance and muscle coordination complexity post-stroke. Journal of Neurophysiology 103, 844–857.

11. Rodriguez K.L., Roemmich R.T., Cam B., Fregly B.J., and Hass C.J. (2013) Persons with Parkinson’s disease exhibit decreased neuromuscular complexity during gait. Clinical Neurophysiology 124, 1390–1397.

12. Fregly B.J., Reinbolt J.A., Rooney K.L., Mitchell K.H., and Chmielewski T.L. (2007) Design of patient-specific gait modifications for knee osteoarthritis rehabilitation. IEEE Transactions on Biomedical Engineering 54, 1687–1695.

13. Fregly B.J., D’Lima D.D., and Colwell C.W. (2009) Effective gait patterns for offloading the medial compartment of the knee. Journal of Orthopaedic Research 27, 1016–1021.

14. Den Otter A.R., Geurts A.C.H., Mulder T., and Duysens J. (2006) Gait recovery is not associated with changes in the temporal patterning of muscle activity during treadmill walking in patients with post-stroke hemiparesis. Clinical Neurophysiology 117, 4–15.

15. Grasso R., Ivanenko Y.P., Zago M., Molinari M., Scivoletto G., Castellano V., Macellari V., and Lacquaniti F. (2004) Distributed plasticity of locomotor pattern generators in spinal cord injured patients. Brain 127, 1019–1034.

16. Meyer A., Patten C., and Fregly B.J. (2013) A surrogate model for lower extremity EMG-to-moment estimation during walking. In: Proceedings of the XXIV Congress of the International Society of Biomechanics. August 4-9, Natal, Brazil.

17. Meyer A.J., Jackson J.N., Patterson M.A., Rao A. V., and Fregly B.J. (2014) Generating three-dimensional dynamically consistent simulations of walking using optimal control. In: Proceedings of the 7th World Congress of Biomechanics. July 6-11, Boston, MA.

18. Erdemir A., McLean S., Herzog W., and van den Bogert A.J. (2007) Model-based estimation of muscle forces exerted during movements. Clinical Biomechanics 22, 131–154.

19. Fregly B.J., Besier T.F., Lloyd D.G., Delp S.L., Banks S.A., Pandy M.G., and D’Lima D.D. (2012) Grand challenge competition to predict in vivo knee loads. Journal of Orthopaedic Research 30, 503–513.

20. Walter J.P., Kinney A.L., Banks S.A., D’Lima D.D., Besier T.F., Lloyd D.G., and Fregly B.J. (2014) Muscle synergies may improve optimization prediction of knee contact forces during walking. Journal of Biomechanical Engineering 136, 021031.

21. Kinney A.L., Besier T.F., D’Lima D.D., and Fregly B.J. (2013) Update on grand challenge competition to predict in vivo knee loads. Journal of Biomechanical Engineering 135, 021012.

22. Pai M., Patten C., and Fregly B.J. (2014) Synergy analysis of pre-treatment EMG differentiates responders and non-responders to stroke rehabilitation. In: Proceedings of the 7th World Congress of Biomechanics. July 6-11, Boston, MA.

23. Zelik K.E., La Scaleia V., Ivanenko Y.P., and Lacquaniti F. (2014) Can modular strategies simplify neural control of multidirectional human locomotion? Journal of Neurophysiology 111, 1686–1702.