Deep Convolutional Bidirectional LSTM Recurrent Neural Network for Epileptic Seizure Detection

Recording the brain electrical activities using Electroencephalogram (EEG) has become the most widely applied tool by physicians for the diagnosis of neurological disorders. In this paper, we propose an automatic epileptic seizure detection system based on raw EEC signals recordings. The proposed system uses a one-dimensional convolutional neural network (CNN) as a preprocessing front-end and a bidirectional long short-term memory (Bi-LSTM) recurrent neural network as a back-end. The system works efficiently on classifying raw EEG signals without the overhead of features extraction. Classification between normal and ictal cases has achieved a 100% accuracy. Using a simple data augmentation technique for the dataset, the classification result between the normal, interictal and ictal cases accomplished a 98.89% average overall accuracy. The evaluation of the proposed system is conducted using k-fold cross-validation to ensure its robustness.