Abstract:
Brain-Computer Interfaces (BCI) is the technology that connects brain signals with external devices. Stroke rehabilitation is one of the most promising Electroencephalogram (EEG)-based BCI applications especially in upper limb stroke rehabilitation. BCI-based rehabilitation is usually done by practicing Motor Imagery (MI). This study developed EEG-based MI hand movement classification method. Hand opening/closing is the basic exercise given to patients in conventional stroke rehabilitation. It is also the movement that has widely been chosen as MI tasks. Wrist flexion/extension and forearm pronation/supination are also the main exercises in conventional rehabilitation. This study evaluated the effectiveness of such movements for MI tasks. Eleven healthy subjects were recruited. Each subject participated in each task respectively. LDA and SVM classifiers gave comparable classification accuracies. For feature selection, Filter Bank Common Spatial Pattern achieved significantly higher accuracies compared to Whole Band Common Spatial Pattern. Session dependent training provided significantly higher accuracies than those of session independent. More training sessions improved subjects MI performance. Moreover, higher number of electrodes gave higher classification accuracy. Considering the accuracy, setup time and the difficulty of setting up EEG headset, the group of nine electrodes would be recommended. The accuracies of classifying each MI task of left hand and right hand also indicate the possibility of classifying EEG data from same side of the brain area.