Tensor Based EEG Feature Extraction And Rehabilitation Application

Nowadays,increasing attention has been paid to the two social issues:the aging of the population and the chronic pressure of young people.As the population aging becomes serious,the number of stroke patients increases accordingly.The young generation easily suffer from chronic stress in the fast pace and competitive society,which leads to the patients of cerebral injury becoming younger.The motor disorder caused by cerebral injury not only brings both psychological and financial stress to the family,but also exerts certain burdens to the society.Fortunately,the "Healthy China" program,which has been emphasized for recent years,indicates that China is paying increasing attentions on health care.This paper proposes a BCI-FES based rehabilitation training system,providing a new approach to motor functional rehabilitation,for stroke patients.Compared with the boring traditional clinical treatment,the active training module and the multimodal feedback module in this system have advantage in encouraging subjects actively participating training and improving the interest of the rehabilitation.In addition,the newly embedded fine motor training module provides a special rehabilitation training for patients with fine motor dysfunction.It is very important to accurately identify the subject’s imagination patterns for our BCI-FES based system which converts motor imagery EEG into commands to control the subjects in virtual environment or external devices.As we know,the performance of some traditional EEG feature extraction method depends on the foreknown active channels and frequency information which is not available about stroke patients.However,almost all of those approaches aim to find one common subspace for projection of all the samples in different classes.Studies have shown that active channels and frequency information are not only subject-dependent but also class-dependent.Thus,in this paper,we propose a tensor-based feature extraction method which attempts to seek individual spatial and spectral subspaces for each class,creating further discriminative information for classification.Finally,we evaluate the effectiveness and robustness of the proposed method on two different datasets including one EEG dataset collected from healthy subjects and one self-collected EEG dataset collected from stroke patients,by comparing with some classical methods.The results demonstrate its superior performance in identifying the motor imagery EEG patterns of stroke patients.