Utilization of Mahalanobis Distance on Motor Imagery EEG Channel Selection

In this study, a Mahalanobis distance based common spatial pattern (CSP)model (MD-CSP) was proposed and its performance applied to brain-computer interface (BCI) and motor imagery EEG signal processing was evaluated. Since EEG signals are often affected by various noises and artifacts, this study suggests a method of using Mahalanobis distance to remove noise and select important EEG channels. The experiment was conducted using the BCI competition IV 2a dataset and compared with CSPRank, L1 Norm of CSP, SCSPrank, FBCSPRank, and E-CSP. As a result of the experiment, MD-CSP performed particularly well at low number of channels and recorded an average accuracy of 60% in the two-channel configuration, outperforming 30%-43% accuracy in established models. MD-CSP showed better accuracy compared to other existing methods and showed a particularly noticeable difference in performance for less than five channels. This study shows that MD-CSP is effective in channel selection and noise removal in BCI system and suggests the possibility of application to real-time portable BCI system in the future.