Epileptic Seizure Detection Based on Feature Extraction and CNN-BiGRU Network with Attention Mechanism

Epilepsy is one of the most widespread neurological disorders of the brain. In this paper, an efficient seizure detection system based on the combination of traditional feature extraction and deep learning model is proposed. Firstly, the wavelet transform is applied to the EEG signals for filtering processing and the subband signals containing the main feature information are selected. Then several EEG features, including statistical, frequency and nonlinear properties of the signals, are extracted. In order to highlight the extracted feature representation of EEG signals and solve the problems of slow convergence speed of model, the extracted features are fed into the proposed CNN-BiGRU deep network model with the attention mechanism for classification. Finally, the output of classification model is further processed by the postprocessing technology to obtain the classification results. This method yielded the average sensitivity of 93.68%, accuracy of 98.35%, and false detection rate of 0.397/h for the 21 patients in the Freiburg EEG dataset. The results demonstrate the superiority of the attention mechanism based CNN-BiGRU network for seizure detection and illustrate its great potential for investigations in seizure detection.