BRAIN COMPUTER INTERFACE - Application of an Adaptive Bi-stage Classifier based on RBF-HMM

Abstract

Brain Computer Interface is an emerging technology that allows new output paths to communicate the users intentions without the use of normal output paths, such as muscles or nerves. In order to obtain their objective, BCI devices make use of classifiers which translate inputs from the users brain signals into commands for external devices. This paper describes an adaptive bi-stage classifier. The first stage is based on Radial Basis Function neural networks, which provides sequences of pre-assignations to the second stage, that it is based on three different Hidden Markov Models, each one trained with pre-assignation sequences from the cognitive activities between classifying. The segment of EEG signal is assigned to the HMMwith the highest probability of generating the pre-assignation sequence. The algorithm is tested with real samples of electroencephalografic signal, from five healthy volunteers using the cross-validation method. The results allow to conclude that it is possible to implement this algorithm in an on-line BCI device. The results also shown the huge dependency of the percentage of the correct classification from the user and the setup parameters of the classifier.