Classifying responses to imagined movements in scalp and intracranial EEG for a Brain Computer Interface

Brain Computer Interfaces (BCI) are based on the analysis of electrical brain signals to produce an accurate identification of a subject's desired action. A 3-class Meta-classification BCI system that allows the classification of different movements or their imagination was developed. It is suitable for signals obtained by scalp and intracranial recordings.;The classification was done in several steps. The first consisted in single-feature classification of each class against the others. This was followed by the selection of the most relevant classifiers and of a method to combine them. The analysis of the different physiological features involved in motor imagery provided the information to tailor the system in a subject-specific way. Meta-classification ensured a more robust performance than single-feature classification, since it takes advantage of the complementarity provided by the different features. A global accuracy of 68.3% was obtained for motor imagery task and 88.4% for real movement tasks.