A study of EEG features for multisubject brain-computer interface classification

BRAIN computer interface (BCI) is a communication technique that aims to detect and identify brain intents and translate them into machine commands to control the operation of electrical and/or mechanical devices. Electroencephalography (EEG) is a widely used imaging technique for noninvasive BCI. Due to EEG non-stationarity, which is typically caused by variation of head size, electrode positions and/or impedance, subjects´ mind states, eye or muscular movements, EEG signals exhibit significant inter-subject variation. As a result, a BCI system trained from a subject may not be directly applicable to others, and a significant amount of time is required to re-calibrate the BCI system to a new subject. This inefficiency is one of the major challenges in EEG-based BCI systems. The goal of this work is to address the multisubject BCI classification by evaluating a set of EEG features and identifying those showing higher stationarity than others.