Performance of a P300-based BCI system improved by a Bayesian single-trial ERP estimation technique

Brain computer interface (BCI) systems based on electroencephalographic (EEG) signals are appealing given their non invasiveness, high temporal resolution, portability and low set- up cost. In particular, P300-based BCI does not require any previous long training of the subject. In this work we assess the improvement of classication performance obtained in a P300-based BCI system by \preprocessing" the signal by a Bayesian ltering procedure for single trial ERP estimation. The reference system is the BCI prototype designed at the IRCSS San Camillo Hospital (Venice, Italy), which embeds a preprocessing procedure based on independent component analysis (ICA). Results from two healthy subjects and four patients aected by amyotrophic lateral sclerosis (ALS) show that classication errors relative to the Bayesian approach for single-trial ERP estimation are at least halved with respect to the reference ICA method.