Post-Stroke Fine Hand Motion Intention Recognition Based on sEMG Decomposition and Residual Spiking Neural Networks.

Area of Science:

• Neuroscience and Biomedical Engineering
• Neuromorphic Computing
• Rehabilitation Technology

Background:

• Fine motor dysfunction in stroke survivors significantly impairs daily activities.
• Effective robot-assisted rehabilitation requires accurate decoding of hand motion intentions from surface electromyography (sEMG).
• Current sEMG-based methods, including those using motor unit spike trains (MUSTs), often underutilize the full potential of sEMG decomposition, especially in post-stroke populations.

Purpose of the Study:

• To propose and evaluate a novel hand motion intention recognition framework integrating sEMG decomposition with a residual spiking neural network (Res-SNN).
• To assess the framework's performance in both neurotypical individuals and stroke survivors.
• To compare the proposed Res-SNN framework against traditional sEMG-based deep residual networks (ResNet) and MUST-based convolutional SNNs (CSNN).

Main Methods:

• Recorded sEMG signals from 14 neurotypical individuals and 7 stroke survivors performing 35 distinct fine hand and wrist movements.
• Developed a framework integrating sEMG signal decomposition with a residual spiking neural network (Res-SNN) for motion intention recognition.
• Evaluated Res-SNN performance, comparing it with ResNet and CSNN across both neurotypical and post-stroke cohorts.

Main Results:

• The proposed Res-SNN framework achieved classification accuracies exceeding 0.95 in both neurotypical and post-stroke cohorts.
• Res-SNN significantly outperformed ResNet in both groups (neurotypical: >0.95 vs. 0.84±0.08; post-stroke: >0.95 vs. 0.90±0.04).
• While comparable to CSNN in neurotypical subjects, Res-SNN substantially outperformed CSNN in stroke survivors (0.95±0.03 vs. 0.71±0.16, P<0.001), demonstrating superior efficacy in this population. The system also exhibited low inference power consumption (5.41 mJ·s).

Conclusions:

• The integration of sEMG decomposition with Res-SNN provides a highly accurate and energy-efficient solution for recognizing hand motion intentions in stroke survivors.
• This approach advances neural decoding technologies and neuromorphic computing applications in human-machine interfaces for rehabilitation.
• The developed framework holds significant promise for improving robot-assisted rehabilitation and enhancing the quality of life for individuals recovering from stroke.