Anxiety Detection with Nonlinear Group Correlation Fusion of Electroencephalogram and Eye Movement

Abstract

Electroencephalogram(EEG) and eye movement have been extensively applied in the detection of anxiety disorders because they can reflect the brain functions and people’s attentional bias. Although our previous work can make good use of the group structure information of EEG and eye movement signals, it mainly models the linear correlation and ignores the nonlinear correlation between two modalities. Therefore, we proposed kernel group sparse canonical correlation analysis (K-GSCCA) to study the nonlinear complex relationship and group structure information among EEG and eye movement features. Firstly, EEG signals were divided into 13 groups according to different brain regions, and eye movement signals were divided into 4 groups according to different visual behaviors. Then, we used the Gaussian kernel function to transform data into kernel space, effectively generated nonlinear cooperative fusion representation. The experimental outcomes demonstrated that K-GSCCA can be effective to solve the nonlinear correlation of group structure information between EEG and eye movement features. Using the support vector machine(SVM) classifier, we finally achieved the best classification accuracy of 87.47% in the fusion of the gamma band of EEG and eye movement features.