Expliciting SSVEP misclassifications with extra-brain activities using time-frequency EEG analysis

Abstract

In order to use brain physiological signals to control a robotic system in the task space, it is mandatory to distinguish as quickly as possible and very reliably bad choices due to wrong brain signals classifications. This allows one to: (i) eventually recover non-desired resulting (robotic) actions, while in the same time, (ii) improve the classifier/controller parameters in order to interpret more precisely the brain signals for the next actions. Instead of using EEG error potential identification (ErrP), we instruct the users to explicit misclassifications using one of the two following extra-brain activities: briefly clenching teeth or closing the eyes. The experiments conducted on three healthy subjects, show that these two extra-brain activities are detectable by EEG time-frequency analysis and in less than one second if the user is focused. Indeed, associated potentials are clearly distinguished after they are made following a bad classification result is revealed to the user. Our analysis and results are based on a brain machine interface (BMI) using the Steady State Visual Evoked Potentials technique (SSVEP).