Graphical models for decoding in BCI visual speller systems

We introduce the use of graphical models in the decoding process of brain-computer interface (BCI) visual speller data. The standard decoding implicitly assumes a simple graphical model which does not incorporate overlap and refractory effects of the brain signals. We propose a more realistic graphical model that does incorporate these effects. The decoding that follows from the graphical model involves the use of multiple classifiers. Our approach is tested on real visual speller data. The results show that the proposed method slightly outperforms the standard decoding method.