Deep learning predicts finger movements from muscle signals after stroke
Deep Neural Networks for Learning Intent from sEMG Signals to Support Hardware Devices for Post-Stroke Neurorehabilitation
Machine Learning
Summary
After a stroke, people often have weak finger movements but their muscles still send signals. The authors worked on reading these muscle signals to figure out which fingers someone intends to move, even if the movement is weak or incomplete. They tested different types of deep learning models on muscle data from stroke patients and found that certain neural network designs work better at decoding finger movement intent. This approach could help create devices that assist stroke patients by understanding their finger movement commands from muscle signals.
What this means in practice
- •For rehabilitation device developers: Develop wearable hardware that decodes five-finger movement intent from a small number of muscle sensors for stroke patient rehabilitation.$Commercial implications: Enables selling compact, wearable neurorehabilitation devices that read finger-specific muscle signals to support post-stroke physical therapy.
- •For embedded systems engineers: Implement efficient neural network models for finger movement decoding in hardware with limited sensors and computational resources.
Authors
Zakariyya Brewster, Divy Wadhwani, Emily Yan, Aidan Wang, Karma Namgyal, Shuting Xie, Markiyan Konyk, Tala Abdelmaguid
Abstract
Finger-specific motor intent is a clinically meaningful control signal for post-stroke neurorehabilitation, where residual muscle activity may remain measurable despite weak or incomplete movement. We study five-finger multilabel intent decoding from impaired-arm high-density surface electromyography (sEMG) in PhysioMio, a bilateral longitudinal dataset collected from stroke patients. A common processing protocol aligns movement labels, applies 20--450 Hz Butterworth filtering and Symlet-4 wavelet denoising, segments overlapping 200 ms windows, and extracts twelve time- and frequency-domain descriptors per channel. Direct LSTM, CNN, and GNN baselines reveal complementary behavior: the LSTM attains the highest subset accuracy (0.545), whereas the GNN attains the highest macro F1 (0.706) and macro AUPRC (0.776). Architecture search then identifies CNN-Large as the strongest single-split CNN, with 0.593 subset accuracy and 0.714 macro F1, while CNN-Micro provides a compact architecture for embedded inference. To match a four-sensor hardware design, we retrain CNN-Micro using channels associated with ECRB, ECRL, FDS, and FDP and exclude the ground electrode from model input. Across five seeds, cross-channel knowledge distillation improves the four-channel student over direct training, reaching $0.5219 \pm 0.0114$ subset accuracy, $0.7612 \pm 0.0038$ finger accuracy, and $0.6095 \pm 0.0058$ macro F1. The selected 123K-parameter model accepts nine windows of 48 features and has been exported to ONNX. These results establish a reproducible software path from post-stroke sEMG to compact five-finger intent prediction for subsequent hardware-in-the-loop evaluation.