Investigating Classification Parameters for Continuous Myoelectrically Controlled Prostheses

Abstract

This work is part of the ongoing research on dexterous and natural  control of upper extremity prostheses using myoelectric signals (MES).  An extensive database of thirty subjects using eight channels of MES  from the right arm was developed. Data were collected from each subject in four sessions on separate days, with each session containing six trials. Each trial consisted of seven limb movements (hand open, hand close, wrist flexion, wrist extension, supination, pronation and rest)  repeated four times, held for three seconds, in a random sequence.  Electrodes were placed on right forearm at muscle sites determined by human physiology of the limb movements and one electrode was placed on the bicep muscle of the right arm. This database will serve as necessary input for the future development and testing of  classification algorithms, including evaluating different classification techniques, comparing different feature sets, and investigating intra- and inter-session variabilities.  In this paper, the effect of channel placement is investigated.  This investigation is performed by comparing classification accuracies using all eight MES channels with various subsets of channels.  Results will provide an indication of which muscle sites are important and how many channels are necessary to maintain a high degree of classification accuracy.