Effect of Spatial Filtering and Channel Selection on Motor Imagery BCI

Abstract

Motor imagery (MI) based BCI has a wide application prospect if a small set of electrodes of EEG are used. In this paper, we compared the effects of different spatial filters, different frequency bands, and different number of channels on the classification accuracy of the EEG of MI using both trained and untrained MI EEG data. Firstly, large Laplacian (LL), common average reference (CAR), and unprocessed raw data were applied respectively. Then the sensorimotor rhythm-based features were extracted by common spatial pattern (CSP), and classified using linear discriminant analysis (LDA) to obtain the classification accuracy with different number of EEG channels involved. The statistic test results show that there is no significant difference in the classification accuracy with different spatial filtering methods. The classification accuracies based on 8, 13, and 22 channels EEG are not significantly different, which indicates that it is possible to use only 8 channels for MI-BCI in application for real-time BCI.