A data driven Information theoretic feature extraction in EEG-based Motor Imagery BCI

Motor Imagery (MI) is the most popular Brain-Computer Interface (BCI) model which aimed at analyzing and classifying the electroencephalogram (EEG) measured without direct human’s motor movements. The EEG recording is measured on the scalp noninvasively, which has nonstationarity and nonlinearity. To tackle the obstacle for analyzing the EEG obtained during MI tasks, we propose a novel feature extraction method by combining the Hilbert-Huang Transform (HHT) and the dispersion entropy (DisEn). Here, we develop the multivariate HHT using intrinsic mode functions (IMFs) obtained through multivariate empirical mode decomposition (MEMD) instead of HHT using existing EMD. By comparing the classification performance with other traditional methods, we validate the improved capacity of the proposed method, which shows its usefulness in MI BCI model.