Characterization of surface electromyography patterns of healthy and incomplete spinal cord injury subjects interacting with an upper-extremity exoskeleton

Rehabilitation exoskeletons may make use of myoelectric control to restore in patients with significant motor impairment following a spinal cord injury (SCI) a sense of volitional control over their limb — a crucial component for recovery of movement. Little investigation has been done into the feasibility of using surface electromyography (sEMG) as an exoskeleton control interface for SCI patients, whose impairment manifests in a highly variable way across the patient population. We have demonstrated that by using only a small subset of features extracted from eight bipolar electrodes recording on the upper arm and forearm muscles, we can achieve high predictive accuracy for the intended direction of motion. Five healthy subjects and two SCI subjects performed voluntary isometric contractions while wearing an exoskeleton for the wrist and elbow joints, generating six distinct single and multi-DoF motions in a total of sixteen possible directions. Using linear discriminant analysis, classification performance was then evaluated using randomly selected holdout test data from the same recording session. Commonalities across subjects, both healthy and SCI, were analyzed at the levels of selected features and the values of commonly selected features. Future work will be to investigate group-specific classification of SCI subjects' intended movements for use in the real-time control of a rehabilitation exoskeleton.