Connectivity pattern modeling of motor imagery EEG

In this paper, the functional connectivity network of motor imagery based on EEG is investigated to understand brain function during motor imagery. In particular, partial directed coherence and directed transfer function measurements are applied to multi-channel EEG data to find out event related connectivity pattern with the direction and strength. The t-test is applied to these connectivity measurements to compare the network between motor imagery and the rest state. The possible relationship between this connectivity pattern and subjects performances are discussed. Based on the Granger causality analysis, a feature extraction method is proposed to compensate for nonstationarity in data. By attenuating the time-lagged correlation, this feature extraction method based on the multi-variate autoregression model is proposed to reduce the effects of noises caused by time propagation. The validity of the proposed method is verified through experimental studies with a two-class dataset, and significant improvement in term of classification accuracy is achieved.