A plantar multi-parameter stimulation-feedback quantitative modeling framework for post-stroke gait rehabilitation

Personalized gait rehabilitation after stroke requires optimization of high-dimensional stimulus-feedback parameters (position, intensity, frequency, pulse width, duty cycle, spatial pattern) with phase-aligned delivery, but most previous studies have used single-factor adjustments and macro-evaluations. Here, we propose SIMStep (Similarity-Driven Personalized Gait Optimization Stimulation-Feedback), a methodological framework that unifies quantification and multi-factor modeling. We implemented a five-factor, five-level design using a dual-modality foot platform (vibration + electricity), with 40 conditions per participant selected via a nearly orthogonal space-filling Latin hypercube. Electromyography (EMG) is recorded from 15 healthy adults. We introduce SGE-DTW, a nonclinical proxy that measures the dynamic time warping similarity between stimulus-evoked EMG and each participant's natural gait EMG (lower values indicate closer to natural gait), enabling cross-subject comparability. A second-order polynomial response surface (validated across subjects and conditions) achieves robust predictive skill value (Skill value (mean ± SD) = 0.85 ± 0.13) and significant overall goodness of fit across subjects (p-values < 0.001), revealing interpretable main and interaction effects (“main effects” are univariate marginal effects after marginalizing over other factors at their trial levels). Notably, the electrical frequency × pulse width interaction consistently shapes the response surface, supporting both individual optima and group-preferred regions. Independent yes–no perceptual judgments of “natural walking sensation” were consistent with the SGE-DTW (Harrell's C = 0.74, 95 % CI 0.68–0.80), providing orthogonal support. Overall, this work should be regarded as a methodological validation in healthy adults—a first step toward, but not yet a direct study in, stroke rehabilitation. Clinical applicability will require stroke-specific normalization and separate validation in dedicated stroke cohorts, where SGE-DTW will be evaluated alongside wearable task-level gait assessments and conventional functional scales.