Using behavioral information to contextualize BCI performance

Abstract

Brain-computer interface (BCI) systems often require millisecondlevel timing precision in order to function reliably. However, as BCI research expands to an ever-widening array of applications, including operation in real-world environments, such timing requirements will need to be relaxed. In addition, overall BCI system design must be improved in order to better disambiguate the numerous, seemingly similar, neural responses that may arise in such environments. We argue that this new area of operational BCI will require the integration of neural data with non-neural contextual variables in order to function reliably. We propose a framework in which non-neural contextual information can be used to better scope the operational BCI problem by indicating windows of time for specific analyses as well as defining probability distributions over these windows. We demonstrate the utility of our framework on a sample data set and provide discussion on many of the factors influencing performance.