Effects of electrode configuration on pattern recognition based finger movement classification

Abstract

Pattern recognition (PR) based myoelectric control could provide intuitive and dexterous control of advanced prostheses. Previous studies showed that the performance of finger movements was not as good as that of wrist movements. As electrode configuration plays an important role in classification performance, this study investigated the effect of the number of electrodes and their locations on finger movement classification. An electrode selection algorithm, sequential forward searching (SFS), was applied on the high density (HD) electrode grid with 192 monopolar electrodes. With the time domain (TD) feature and linear discriminant analysis (LDA), it was found that the error rate was dramatically decreased with the number of electrodes increasing from one to ten. Under the optimized electrode configuration, the error rate could be lower than 10% with 8 electrodes, and 5% with 18 electrodes. The importance of the electrodes was measured and the results showed that the effective site for classification was mainly located around the flexor digitorum superficialis and extensor digitorum communis. This study provides the guideline for optimal placement of electrodes in finger movement recognition, and potentially provide sufficient controllability of advance prostheses with individually articulated fingers.