Closed Loop Brain Computer Interface Based on Improved Common Spatial Patterns and Visual Feedback

Abstract

In order to improve the active participation of patients with stroke or nerve injury in the lower limb rehabilitation training, an on-line closed loop brain computer interface is designed based on human lower limb motor imagery and visual feedback. And an improved common spatial pattern algorithm, which based on the correntropy induced metric and sub-frequency band analysis, is established to improve the recognition rate of the motor intention. Due to the low signal noise ratio and classification accuracy of the motor imagery EEG signals, correntropy induced metric is adopted to improve the objective function of the traditional common spatial patterns (CSP). The distance term of the objective function can be adjusted dynamically to alleviate the negative effects of noise. Because the different frequency band signals have different information, nine sub-frequency bandpass filters are used to filter the signal. And the features extracted from each sub-band signal are fused. Therefore, the improved common spatial pattern algorithm based on the correntropy induced metric and sub-frequency band analysis is established. Then, based on the brain control experiment paradigm of human lower limb motor imagery, EEG data of lower limbs motor imagery (idle, foot and leg) are collected. Support vector machine (SVM) is optimized as a classification model for the motor imagery. Based on the study above, a brain computer interface based on improved common spatial pattern algorithm and SVM is built. When participant images the movements, the user’s visual feedback is given to the user through the body movements of the virtual character in the virtual reality scene, and a closed loop brain computer interaction system is constructed. Experiments verified the effectiveness of the improved common space algorithm and the feasibility of closed-loop brain computer interface, and the closed loop interaction between the brain and computer is achieved initially.