Preliminary Investigation of Myoelectric Control of an Assistive Neck Exoskeleton by Individuals With Amyotrophic Lateral Sclerosis.

Neck weakness limits head control and quality of life for individuals with Amyotrophic Lateral Sclerosis (ALS). The Utah Neck Exoskeleton can restore neck motion, but current control methods-joystick and gaze tracking-have limited accessibility and reliability. These preliminary offline analyses investigate neck electromyography (EMG) as an alternative control modality from ALS patients. EMG signals were recorded from four male participants with ALS while performing neck flexion/extension, axial rotation, and lateral deviation. The resulting dataset was used to train convolutional neural networks (CNNs) per patient to classify either head position or movement direction from EMG features offline. Position classification significantly outperformed direction classification, with a mean accuracy of 82.5% $\pm ~0.010$ across participants. Performance was consistent when controlling one, two, or all three neck degrees of freedom. A subset of participants with sufficient residual motor function also completed neck movements while talking or chewing. Classification accuracy decreased during talking and chewing, although these effects were not statistically significant. Importantly however, training CNNs with diverse data that included periods of talking and chewing improved algorithm robustness across all conditions. These findings suggest that neck EMG signals can reliably predict intended head movements in ALS, even in the presence of weak and often confounding muscle activity. Offline accuracy and real-time computational speed suggest the approach is feasible for future online user-in-the-loop studies. Altogether, this pilot work advances EMG-based assistive technology for individuals with severe motor impairments, laying the groundwork for clinically viable, intuitive control systems.