Classifying EEG motor imagery signals using supervised projection pursuit for artefact removal

A novel bottom-up approach for EEG signal artefact removal and classification is presented. An eigenbrain is constructed from two eigenchannels on a per subject basis. An eigenchannel provides a characteristic vector space to discriminate motor imagery from resting state signals in both the µ and β frequency bands. Discrimination between left and right hand motor imagery is achieved at the eigenbrain level as a direct sum of the eigenchannel vector spaces. It was found this methodology is viable using only 5% of available motor imagery trials during training, yielding 92.6% classification accuracy on the selected test subject. Furthermore, the utility for real-time applications is promising with a rapid classification that takes less than 100 ms after the initial calibration procedure is completed.