Muscle synergy-driven ensemble learning framework for individualized stroke gait rehabilitation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Stroke survivors often exhibit gait abnormalities due to neuromuscular impairments.
• Muscle synergy analysis provides insights into motor control deficits.
• Current clinical decision-making in rehabilitation can be enhanced by objective, data-driven tools.

Purpose of the Study:

• To develop and validate a novel ensemble machine learning (ML) framework for stroke gait rehabilitation.
• To integrate neurophysiological principles of muscle synergy analysis into an ML model for improved clinical decision support.
• To enhance the interpretability and classification performance of ML models in identifying stroke-related neuromuscular impairments.

Main Methods:

• Extracted muscle synergies using non-negative matrix factorization from surface EMG data of 380 participants (120 healthy, 260 post-stroke).
• Utilized a bidirectional decomposition process to derive feature vectors capturing spatial (W) and temporal (H) synergy deviations from normative patterns.
• Employed a hierarchical ensemble model with meta-regression integrating separate classifiers for spatial and temporal features, validated using SHAP values for interpretability.

Main Results:

• The ensemble ML framework achieved classification accuracies exceeding 98% on the internal test dataset.
• Feature importance analysis confirmed the clinical relevance of learned classification criteria.
• SHAP values provided sample-specific explanations, ensuring individual-level interpretability of predictions regarding neuromuscular impairment.

Conclusions:

• The proposed framework effectively integrates neurophysiological principles with ML for stroke gait rehabilitation.
• The model demonstrates high accuracy and interpretability in identifying stroke-related neuromuscular impairments.
• This approach lays the groundwork for incorporating neurophysiologically grounded ML models into clinical practice for enhanced rehabilitation outcomes.