Discriminating multiple motor imageries of human hands using EEG

We investigated the feasibility of discriminating four different motor imagery (MI) types from both hands using electroencephalography (EEG) through exploring underlying features related to MIs of thumb and fist from one hand. New spectral and spatial features related to different MIs were extracted using principal component analysis (PCA) and squared cross correlation (R
²
). Extracted features were evaluated using a linear discriminant analysis (LDA) classifier, resulting in an average decoding accuracy about 50%, which is significantly higher than the guess level and the 95% confidence level of guess. The preliminary results demonstrate the great potential of extracting features from different MIs from same hands to generate control signals with more degrees of freedom (DOF) for non-invasive brain-computer interface applications. In addition, for movement related applications, especially for neuroprosthesis, the present study may facilitate the development of a non-invasive BCI, which is highly intuitive and based on users' spontaneous intentions.