Identification of EEG Features in Stroke Patients based on Common Spatial Pattern and Sparse Representation Classification

Xuefeng Lei1, Luyun Wang1, Wanzeng Kong1, Yong Peng2, Sanqing Hu1, Hong Zeng1, Guojun Dai1, Ruoyu Jin1, Junfeng Sun2, Shanbao Tong2

  • 1Hangzhou Dianzi University
  • 2Shanghai Jiao Tong University

Details

11:30 - 13:30 | Fri 26 May | Emerald III, Rose, Narcissus & Jasmine | FrPS1T1.31

Session: Poster I

Abstract

Electroencephalography (EEG) and brain-computer interfaces (BCI) are receiving increasing attention and expanding application in stroke study. To identify stroke patients and normal controls during mental rotation task, common spatial pattern (CSP) algorithm is employed to extract features from binary-class EEG which will be further to form the dictionary for sparse representation. In the classification process, sparse representation-based classification (SRC) method is used; specifically, each test trial is sparsely represented over the formed dictionary and the sparse coefficient is obtained by solving a l1-norm regularized least squares minimization objective. A series of experiments demonstrated the effectiveness of features extracted by CSP of both classes and the SRC could obtain excellent results in classification. These results suggest that stroke patients have distinct EEG feature from normal controls. These EEG features may be potential biomarkers to monitor the rehabilitation process of stroke patients.