Discrimination of Four Classes in Brain Computer Interface Based on Motor Imagery

This study investigated the classification of multiclass motor imagery for electroencephalogram (EEG)-based Brain-Computer Interface (BCI) using independent component analysis (ICA), principle component analysis (PCA) and support vector machine (SVM) techniques. The dataset used is available on BCI competition IV that contains EEG signals for 9 subjects who performed left hand, right hand, foot and tongue motor imageries (MI). The ICA technique appears well suited for performing source separation in domains where the number of independent signal sources is equal to the number of electrodes or sensors, which is not applicable in the case of EEG sources, since we have no idea about the effective number of statistically independent brain signals related to the EEG recorded from the scalp, also we proved that right hand can activate the same areas of left hand in the brain, while foot can activate the same areas of hands and tongue. Thus we did not have high expectations for separating the same signal sources in all sessions and this justify the overall accuracy of 33+/-2% that we got when using the combination of ICA and SVM techniques.