EMG-based motion intention detection for control of a shoulder neuroprosthesis

Abstract

A method for predicting shoulder and motions from electromyograms (EMGs) from shoulder muscles using a time-delayed artificial neural network (TDANN) is described. The chosen network was found to be capable of characterizing the nonlinear and dynamic relationship between the EMG signals recorded from 6 shoulder muscles and the resulting shoulder and elbow motions in 5 able-bodied subjects. Preliminary work in one individual with tetraplegia due to spinal cord injury indicate that the same TDANN structure (although with a different set of muscle EMGs) will be also be sufficient to detect these motions in this population. This ability to detect shoulder and elbow motions would allow neuroprostheses based on functional neuromuscular stimulation (FNS) to appropriately vary stimulation patterns in a very natural manner for different tasks.