Selection of EEG channels based on Spatial filter weights

Most of the Electroencephalography (EEG) based brain computer interfaces (BCIs) use large number of channels to capture the signals from subject's brain. This is one of the major issues in commercial and out of the lab usage of such systems. Common Spatial Pattern (CSP) algorithms are widely used for feature extraction in BCI systems for motor imagery. As the EEG signals have noise and overfitting issues, various regularized CSP algorithms are introduced to overcome these factors. In this work, a method is introduced to find the variant of CSP that achieves maximum classification accuracy with least number of EEG channels. The approach is based on firstly identify the spatial filter weights using complete set of channels. In the next step, the channels are selected based on the maximal filter weights. Channels with highest values are included to find the accuracy in next run and ultimately the optimum combination of number of channels, CSP variant and classification accuracy are reported. Multichannel data comprised of 60 electrodes from BCI Competition III dataset IIIa from three subjects who performed left hand, right hand, foot and tongue MI is considered for this purpose. For the data analyzed in this study, it is found that instead of using data from 60 electrodes only six can be used without significantly compromising the classification accuracy.