Short Time EEG Connectivity Features to Support Interpretability of MI Discrimination

Abstract

Brain connectivity analysis during motor imagery (MI) tasks has evolved as an essential and promising tool for its use in brain-computer interfaces (BCI). Many approaches devoted to BCI systems focus on the distinction between different MI tasks from electroencephalogram (EEG) signals. However, given the non-stationarity of the brain activity, the MI discrimination yields to different classification performances between subjects. Here, we introduced an MI discrimination system from EEG signals to reveal relevant brain connectivity patterns associated with a specific MI protocol. Indeed, we employ a windowed-based feature representation using the well-known Common Spatial Pattern (CSP) technique. Then, the classification performance along temporal windows is related to a Phase Locking Value (PLV)-based connectivity measure. Obtained results show a remarkable relationship between high classification performances and the subject coupling with the acquisition protocol concerning the windows that present the MI stimulus.