Independent component analysis using clustering on motor imagery EEG

Motor imagery is a popular paradigm in electrophysiology research and brain computer interface but the evoked EEG signals always contaminated significantly. In this paper we use the Independent Component Analysis to enhance the signal-to-noise ratio of multi trail EEG signals evoked by imaginary hand movement. Infomax algorithm was used to decompose multi channel EEG signals into independent components trail by trail, and then an automatic clustering method was used to group these components into several clusters. For the higher similarity between task relevant components, they can be assembled into one cluster that occupies the highest mean mutual information of pairwise components intra cluster. Furthermore, the reconstructed signals of task relevant cluster showed a high discrepancy features to left versus right hand task, which evaluated by Fisher criterion scores and served as the signal-to-noise ratio measurement.