DSLWNet: a dual-stream lightweight deep learning network for the detection of Epileptic seizures using EEG signals

Epilepsy is one of the most prevalent neurological disorders, and the accurate detection of epileptic seizures is challenging. Therefore, a dual-stream deep learning network is proposed in this research to extract the deep features by utilising the scalograms and time-series EEG signals. In this, first, the convolutional neural network (CNN) extracts the spatial dimensional features from the scalogram images, and then the squeeze and excitation (SE) technique enhances the relevant informative features by adjusting the channel weights. Correspondingly, the gated recurrent unit (GRU) extracts the temporal characteristics from the time-series EEG signal, and then for assigning more weights to the significant features the confined attention (CA) mechanism is included. Next, the extracted features are fused to form a deep feature set for the accurate detection of seizures using the support vector machine (SVM) classifier. Further, to improve the seizure detection rate, the regression at the end of variational mode extraction techniques (VME) is employed in the preprocessing stage. In addition, the performance of the proposed dual-stream lightweight seizure network (DSLWNet) is evaluated using the CHB-MIT and Bonn datasets. The experimental outcomes show the superiority of the proposed work in seizure detection by achieving an accuracy of 98.67% and 99.5%, respectively.