Optimizing spatio-spectral filters by motor imagery pattern quantification in self-paced Brain Computer Interface

Analyzing ongoing brain activities and distinguishing no control (NC) state of users are the most challenging parts in the self-paced BCI. Many spatial filters such as Common Spatial Pattern and its other developed versions have been proposed to differentiate among specific motor imagery (MI) activities, but they didn´t result in significant achievements in the self-paced BCI. To overcome these drawbacks, this paper proposes a new spatio-spatial filter optimization method by a novel quantification measure for motor imagery patterns. Maximizing the correlation between the linear mixtures of motor cortex channels and motor imagery patterns is used as goal function for genetic optimization algorithm. No sensitivity to initial value is the significant property of this evolutionary algorithm. The most important consequence of this method is increasing the resolution of motor imagery pattern and also improving the motor imagery detection rate in self-paced BCI. Our approach was validated on self-paced dataset 1 of the BCI Competition IV and was compared to different spatial filters including CSP, TRCSP, WTRCSP, and Laplacian filter. The proposed method achieved the highest Area under ROC among the other methods.