HD-sEMG-Based Control Using Neck Muscles and Shallow Neural Networks: Assessing Performance in Rehabilitation-Oriented Tasks.

This paper investigated the suitability of the integrated Recursive Rehabilitation Control Network (RRC-Net)/High-Density Electrode Array (HDE-Array) system for performing two multi-Degree of Freedom (DoF) control tasks, developed as proxies for Functional Electrical Stimulation control: 1) a cursor-based task; and 2) a 3-DoF hand kinematic model control task. The goal of this study is enhancing rehabilitation independence for individuals with spinal cord injuries. The system was validated on both healthy and tetraplegic subjects. The hypotheses that users could successfully perform these tasks using the system and that there would be no significant performance differences between healthy and tetraplegic participants were assessed. The experiment involved 10 tetraplegic and 8 healthy subjects who completed a training phase followed by two testing phases. High-Density surface Electromyography (HD-sEMG) signals recorded from the neck during the training phase were used to train RRC-Net, a neural network designed to estimate multi-DoF movements. Subjects then performed the two control tasks in the testing phase, and performance metrics were analysed and compared between groups. Healthy and tetraplegic subjects achieved high performance in both control tasks. Hand position control performance between the two groups presented no statistically significant differences in Mean Global Distance (MGD) ( ${p} =0.93$ ) or Mean Angular Distance (MAD) ( ${p} =0.77$ ). Similarly, cursor control task performance showed no significant differences in Task Completion Score (TCS) ( ${p} =0.68$ ) or Normalised Distance (ND) ( ${p} =0.63$ ). The system's simplicity, comfort, and effectiveness highlight its potential for rehabilitation, providing a non-invasive method for controlling assistive devices.