Wrist Torque Estimation via Electromyographic Motor Unit Decomposition and Image Reconstruction

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Non-invasive neural interfaces leverage decomposed motor units (MUs) from surface electromyography (sEMG) for human-machine interaction.
• Existing MU-based methods often rely solely on discharge rate (DR), potentially missing crucial local signal information and MU interactions.

Purpose of the Study:

• To propose and evaluate an MU-specific image-based scheme for enhanced wrist torque estimation.
• To improve the accuracy and robustness of neural decoding for intuitive human-machine interfaces.

Main Methods:

• High-density sEMG signals were decomposed into motor unit spike trains (MUSTs).
• MU-specific images were reconstructed using MUSTs and motor unit action potentials (MUAPs).
• A convolutional neural network (CNN) was employed to extract features from these images for torque estimation.

Main Results:

• The proposed image-based method demonstrated superior performance over conventional DR-based and deep-learning regression approaches.
• Achieved high estimation accuracy with R² values of 0.82 ± 0.09 and 0.89 ± 0.06, and nRMSE of 12.6 ± 2.5% and 11.0 ± 3.1% for different wrist movements.
• Features extracted from MU-specific images showed a stronger correlation with recorded torques than DR.

Conclusions:

• The MU-specific image-based scheme offers a promising advancement for non-invasive neural interfacing and intuitive control.
• This method effectively captures richer information than DR, leading to more accurate wrist torque estimation.
• The findings pave the way for more sophisticated and responsive human-machine interaction systems.