DWT-Net: Seizure Detection System with Structured EEG Montage and Multiple Feature Extractor in Convolution Neural Network

Figure 5

Systematic flow of the proposed feature extraction method with 5-second window. Firstly, the EEG signals from TUSZ dataset are restructured into the proposed montage of 19 channels. Secondly, 5-second windows of epileptic and normal EEG are extracted according to the labeling file of the sessions. The windows then go through a Butterworth filter to eliminate power-line interference. After -score normalization, the signal is decomposed into discrete wavelet coefficients. Four coefficient packages representing information of lower frequency bands are used as input of the classifier.