A Data-Driven Investigation on Surface Electromyography Based Clinical Assessment in Chronic Stroke

Area of Science:

• Biomedical Engineering
• Neurorehabilitation
• Machine Learning in Healthcare

Background:

• Robot-assisted rehabilitation utilizing surface electromyography (sEMG) aids chronic stroke survivors in regaining upper limb function.
• Current evaluation methods rely on traditional manual assessments, lacking objective, real-time feedback.
• A need exists for automated assessment tools to enhance rehabilitation efficacy.

Purpose of the Study:

• To develop a novel data-driven model using sEMG signals for automated assessment of upper limb motor function.
• To map sEMG data to established clinical scales like the Fugl-Meyer Assessment (FMA) and Modified Ashworth Scale (MAS).
• To evaluate the model's performance against manual assessments in chronic stroke survivors.

Main Methods:

• A three-layer backpropagation neural network (BPNN) was constructed to process sEMG features (MAV, ZC, SSC, RMS) from four upper limb muscles.
• Twenty-nine chronic stroke participants underwent 20 sessions of sEMG-driven robot-assisted upper limb rehabilitation.
• sEMG data and manual FMA/MAS scores were collected pre- and post-intervention for model training and validation.

Main Results:

• The BPNN model demonstrated high accuracy in mapping sEMG data to FMA scores (r > 0.9, P < 0.001).
• Significant correlations were observed between mapped and manual FMA subscores (FMA-wrist/hand, FMA-shoulder/elbow) pre- and post-intervention.
• The model also showed strong correlations with manual MAS scores for fingers, wrist, and elbow post-intervention (r = 0.91, 0.88, 0.91).

Conclusions:

• A successful sEMG data-driven BPNN model was developed for automated assessment of upper limb motor function in chronic stroke.
• The model shows significant potential for objective, automated assessment in post-stroke rehabilitation.
• Further validation with larger sample sizes is recommended for broader clinical application.