State-Space Modeling of Real-Time Visual Biofeedback and Late-Stance Belt-Speed Modulation for Quantifying Error- and...

Area of Science:

• Neurorehabilitation
• Biomechanics
• Computational modeling

Background:

• Stroke survivors often exhibit impaired lower limb propulsion.
• Restoring gait function is crucial for mobility and independence.
• Current rehabilitation strategies can be enhanced by understanding underlying learning mechanisms.

Purpose of the Study:

• To develop a model-constrained state-space model to analyze gait improvements after stroke.
• To investigate the association between real-time visual biofeedback (VB) and phase-specific belt modulation (PF) with enhanced limb propulsion.
• To characterize the contributions of different learning processes (error-based, use-dependent) to propulsion gains.

Main Methods:

• Employed an exploratory, model-constrained state-space model on data from 12 chronic stroke participants.
• Utilized the Adaptive Propulsion Enhancement eXperience (APEX) system with an instrumented split-belt treadmill.
• Modeled step-to-step ground reaction forces using an extended Kalman filter and smoother.

Main Results:

• The model accurately reconstructed within-session propulsion trajectories (R² = 0.976).
• Propulsion gains were primarily attributed to error-based learning (EBL, ~56%) and use-dependent learning (UDL, ~43%), with minimal direct feed-through (~1%).
• Within EBL, visual biofeedback and belt modulation contributed comparably, indicating adaptive processes drove improvements.

Conclusions:

• Within-session propulsion enhancement post-stroke is linked more to adaptive learning than direct feedback.
• The developed model provides interpretable parameters for understanding within-session gait learning.
• This framework supports hypothesis generation for personalized, mechanism-informed post-stroke rehabilitation strategies.