Using Phase Synchronization to Improve the Performance of Spatial Filter during Motor Imagery EEG Classification

Abstract

Motor imagery-EEG based on brain computer interface (BCI) provide a promising way to establish pathways for neural center and external equipment, which has been widely used in the fields of neurological rehabilitation and robot control. During this process, the performance of BCI depends on the accuracy of decoding for the motor intention. Due to the high efficiency and simplicity, spatial filtering is often used to extract the amplitude feature for MI classification. However, changes of physiological condition for subjects would cause high variability of EEG signal within the trials/sessions, which brought huge challenge of feature extraction and classification. The state of phase synchronization for EEG is an important information to evaluate the motor intention, which is rarely combine with amplitude feature to realize the pattern recognition. Therefore, in this study, we propose a model integrating phase and amplitude information for binary classification in MI task. Firstly, we adopt the phase-locked values to calculate the time section with degree of phase synchronization that contains the maximum discriminant information. Then, common spatial pattern was performed to extract amplitude feature for α and β bands integrating the distribution of phase information. Next, SVM was utilized to classify feature vectors and realize the binary MI decoding. Five subjects recruited to participate in the experiment and results show that the information of phase synchronization significantly improves the classification performance of spatial filter.