Deep Learning Methods for Seizure Prediction by Evaluating EEG Signals

Epilepsy is one of the most common central nervous system diseases. It is characterized by seizures between different periods of time and may endanger the patient's life if they are in a state of work or in a critical place or situation. Anticipating the coming of seizures is important in protecting the patient's life, as they will take some measures or take medications to reduce the risks or prevent the seizure from occurring. In this paper, two deep learning methods are suggested and evaluated; the first model is based on a Convolutional Neural Network (CNN), and the second model is based on Gated Recurrent Unit- Long Short-Term Memory (GRU-LSTM). The models are used in order to distinguish between the preictal state and interictal state, and then predict the onset of the seizure. The methods were based on patient comfort while using the device, using 5 electrodes of Electroencephalogram (EEG) signals and reducing the alert period to only 10 minutes before the onset of the seizure. Results of 90% in terms of accuracy and 90% in terms of sensitivity were obtained using the first model. In addition, 71% accuracy and 73% sensitivity were obtained using the second model. Finally, with respect to the second method by using the average voting technique, a portion of the EEG signal with a length of 2 minutes was used, and the obtained results were 74% accuracy, and 84% sensitivity.