A Novel Feature Extraction Method for Motor Imagery BCI

Abstract

Imagination of limb movement is reflected in EEG signals, which is called motor imagery (MI). MI can be used in brain-computer interface (BCI) applications. In this paper, a new feature extraction method is proposed for MI-based BCI. A Gaussian spatial filter is used in the pre-processing stage, to map the effect of brain sources on the electrodes. Enhanced signals, obtained by preprocessing, are decomposed into standard frequency bands. A practical BCI system should be simple and fast, as much as possible. Therefore, to reduce the computational cost, signals of each frequency band are fed to the common spatial pattern (CSP) block for channel selection. In this paper, a blind source separation (BSS) based technique is proposed to improve feature extraction. As a result, the learning quality of the BCI system has been increased. To assess the proposed BCI system, it is applied to dataset IVa of BCI competition III. The average values of accuracy, sensitivity, specificity, Mathew's correlation coefficient, and F1 score on five subjects for two MI-tasks, are 990%, 98%, 99%, 97%, and 99%, respectively. Results indicate satisfactory performance of the proposed approach.