Feasibility of Error-Related Potential Detection as Novelty Detection Problem in P300 Mind Spelling

In this paper, we report on the feasibility of the Error-Related Potential (ErrP) integration in a particular type of Brain-Computer Interface (BCI) called the P300 Mind Speller. With the latter, the subject can type text only by means of his/her brain activity without having to rely on speech or muscular activity. Hereto, electroencephalography (EEG) signals are recorded from the subject's scalp. But, as with any BCI paradigm, decoding mistakes occur, and when they do, an EEG potential is evoked, known as the Error-Related Potential (ErrP), locked to the subject's realization of the mistake. When the BCI would be able to also detect the ErrP, the last typed character could be automatically corrected. However, since the P300 Mind Speller is optimized to correctly operate in the first place, we have much less ErrP's than responses to correctly typed characters. In fact, exactly because it is supposed to be a rare phenomenon, we advocate that ErrP detection can be treated as a novelty detection problem. We consider in this paper different one-class classification algorithms based on novelty detection together with a correction algorithm for the P300 Mind Speller.