Finger Movement Classification for an Electrocorticographic BCI

We study the problem of distinguishing between individual finger movements of one hand using electrocorticographic (ECOG) signals. In previous work, we have shown that ECOG signals have high predictive accuracy and spatial resolution for classifying hand versus tongue movements. In this paper, we significantly extend this paradigm by studying the first 5-class classification problem for ECOG, and show that an average 5-class accuracy of 23% across 6 subjects is possible using as little as 10min of training data. In addition to opening up possibilities for higher-bandwidth brain-computer interfaces, the use of finger movements for control may yield a more intuitive mapping from ECOG signals to control of a prosthetic. Although this study uses real movements, our results provide the foundation for understanding ECOG signal changes during finger movement.