Detecting Intentional Mental Transitions in an Asynchronous BCI

The inclusion of mental tasks transitions detection (MTTD) has proven a useful tool in guiding the transduction process of a BCI working under an asynchronous protocol. MTTD allows for the extraction of the signal's contextual information in order to infer the user's intentionality at a given moment and thus correcting possible classification errors. Despite the good results shown, the algorithm previously proposed \cite{1} does not show good behavior in contexts where the user gets online feedback. The algorithm that we propose in this paper, like its antecessor, is based on canonical variates transformation (CVT) and on distance-based discriminant analysis (DBDA), but it has a new transitions detector based on Kalman filtering. In addition, it includes a classifier supervisor based on heuristics rules that exploit transition detection as well as inconsistencies between subject's mental intention and the associated EEG. These heuristic rules lead to significant improvements of the BCI in terms of both classification accuracy and channel capacity, adapting itself to the user's needs.