Automatic Seizure Detection Based on a Novel Multi-feature Fusion Method and EMD

Abstract

The analysis of electroencephalogram (EEG) signals is becoming more important because of the time-consuming and large bias of traditional visual detection technology, especially in diagnosis of epilepsy. Based on the nonlinear and non-stationary of the EEGs, empirical mode decomposition (EMD) is applied to decompose the original signals into intrinsic mode functions (IMFs). In this paper, after getting the IMFs, the fluctuation index and variation coefficient are calculated to analyze the amplitude change of IMFs. In order to better reflect the information of EEG signals, Hilbert transform is applied to obtain the instantaneous frequency for each IMF. Then, the novel feature named fluctuation index and variation coefficient of instantaneous frequency for IMFs are calculated. Furthermore, feature based on sample entropy of the first order difference is extracted. Finally, both of the calculated features will put together as a fusion feature into SVM for classification. The proposed method is evaluated using the Boon epileptic dataset and the highest average classification accuracy is 99.59%, showing a powerful method to detect seizure.