Twofold classification of motor imagery using common spatial pattern

Motor imagery (MI) is a mental rehearsal of movement without any body movement. Brain-Computer Interface (BCI) uses MI in the neurological rehabilitation, especially in stroke rehabilitation to restore the patient's motor abilities. BCI based on MI translates the subjects motor intent into control signals to control the devices like robotic arms, wheelchairs or to navigate the virtual worlds. In this work, multichannel electroencephalogram (EEG) signals of imagination of a right hand and right foot movement is considered. Common spatial pattern (CSP) is used to estimate the spatial filters for the multi-channel EEG data. The spatial filters lead to weighting of the channel/electrodes according to their variance in discriminating the two tasks performed. Channels with the largest variance are considered as significant channels. A two-fold classification method using support vector machine (SVM) is used to classify the test signal into right hand movement and right foot movement. In the present work, the analysis conducted demonstrate that the proposed twofold classification scheme can achieve upto 94.2% of accuracy in discrimination of the two tasks performed. The high-recognition rate and computational simplicity make CSP a promising method for an EEG-based BCI.