Optimization of BCI Speller Based on P300 Potential

Abstract

We report our studies on a Brain Computer Interface (BCI) speller application with an aim to optimize its performance and usability. We study the dependence of the spelling accuracy as a function of (a) the number of visual stimuli (repetitions) presented to the user, (b) the P300 segment length used, (c) the number of channels used, and (d) the amount of data used in training, in terms of the number of characters and repetitions. Reducing the number of repetitions results in a direct reduction of the time needed to spell a character, while minimizing the number of channels translates to shorter subject preparation time and thus improves the usability of the system. The usability is further enhanced by decreasing the training required, while maintaining the accuarcy. We show that very high accuracies of the order of 99% can be achieved with a short training session of less than 10 minutes using only about 10 channels. The high accuracies, short training and preparation time requirements along with real-time performance make this BCI speller a viable communication tool for severely disabled individuals, who have no other means to communicate with the external world.