The ability to predict who will benefit from which treatment and how to customize treatments to individuals is a grand challenge in the biomechanics community. In particular, due to the heterogeneity of the post-stroke population, selection of optimal rehabilitation protocols that maximize recovery of walking function often involves extensive trial and error, resulting in patient fatigue and increased healthcare costs. Treadmill-based gait training is commonly prescribed for post-stroke rehabilitation but yields mixed responses possibly due to the use of standardized approaches instead of customized protocols. The overall purpose of this dissertation was to understand and predict user responses to various treadmill training protocols to determine who will benefit from which protocol. This goal was accomplished through a combination of musculoskeletal modeling and simulation techniques and experimental gait analysis.

In the first Aim we developed and evaluated the first predictive framework to estimate several common gait adaptations on a fixed-speed treadmill at a range of belt speeds. This proved the feasibility of combining musculoskeletal models and direct collocation optimal control methods to simulate gait on a treadmill. Although fixed-speed treadmill gait training can be a useful rehabilitation tool, adaptive treadmill control may provide a more similar training environment to community ambulation.

In Aim 2 we sought to create a customizable adaptive treadmill controller to promote increased user propulsion, a common goal of post-stroke gait rehabilitation. We recruited and tested a cohort of 22 young healthy adults and the results show that we can customize an adaptive treadmill to target increased propulsion. To our knowledge this was the first experiment to demonstrate that an adaptive treadmill controller combining propulsion, spatiotemporal, and position-based control can be customized to target therapeutic outcomes such as increased propulsive force.

Building on these experimental results, in Aim 3 we implemented the adaptive treadmill in the simulation framework by modifying the cost function to enable differential gains for propulsive impulse, step length, effort, and speed. We found that increasing the weight on the propulsive term in the simulated adaptive treadmill resulted in enhanced propulsion.

Finally, in Aim 4 we incorporated unilateral weakness into the model with adaptive treadmill control to investigate responses of individuals with different poststroke impairment levels. The predictive simulation results suggest that our modifications may improve the efficacy of the adaptive treadmill as a rehabilitation tool to increase paretic propulsion for stroke survivors with mild, moderate, and severe hemiparesis.

Overall, the results of this dissertation represent an important step toward a comprehensive approach to designing custom treadmill-based therapeutic interventions for post-stroke gait rehabilitation. Future work should seek to further customize adaptive treadmill control through a human-in-the-loop optimization experiment and patient-specific models of stroke impairments.