Spatially sparse spatio-spectral filters for feature extraction in BMI applications

The common spatio-spectral pattern method is an extension of the traditional common spatial pattern technique that combines spectral filtering with the original spatial filtering. All recording channels are combined when extracting the variance as input features for a brain machine interface. This results in overfitting and robustness problems of the constructed system in presence of high number of channels. Here, we construct spatially sparse common spatio-spectral pattern method in which only a subset of all available channels is linearly combined when extracting the features. We utilized a modified version of the recently introduced recursive weight elimination technique to select a subset of electrodes for spatio-spectral projections. We evaluate the performance of the proposed method to distinguish between the movements of the first three fingers of the hand using electrocorticogram signals of the brain computer interfaces competition 2005. We observed that spatially sparse spatio-spectral filter outperforms both original common spatial pattern and non-sparse spatio-spectral filter and results in improved generalization in classification.