Conclusions
In spite of unequal power distribution, a separable MI bispectral features could be found resulting into nearly equal (69%) left and right MI accuracy using the BSP features in all sessions (120 trials). This happens due to inherent properties of BSP and not usual with power spectrum based BCI. Also the higher CA comes if both mu and beta bands of EEG is considered. This indicates that the post-stroke subject’s MI related EEG contains modulated mu and beta rhythms. Compared to the rest state, CA increases during MI time course (Fig. 1). But, the post-stroke subjects perform delayed (6-7s) MI, peaking closer to the end of the trial. Hence it is concluded that the BSP based BCI can be effective for stroke rehabilitation.