Epileptic Seizure Prediction from Spectral, Temporal, and Spatial Features of EEG Signals Using Deep Learning Algorithms

Epilepsy is one of the most common brain disorders that greatly affect patients life. However, early detection of seizure attacks can significantly improve their quality of life. In this study, we evaluated a deep neural network to learn robust features from electroencephalography (EEG) signals to automatically detect and predict seizure attacks.

The architecture consists of convolutional neural networks and long short-term memory networks. It is designed to simultaneously capture spectral, temporal, and spatial information. Moreover, the architecture does not rely on explicit channel selection algorithms. The method is applied to the Childrenchr('39')s Hospital of Boston-Massachusetts Institute of Technology dataset (CHB-MIT). To evaluate the method, the proposed model is trained in the patient-specific approach.

The proposed architecture achieves a sensitivity of 90.7 ± 7.9 percent, a false prediction rate of 0.12/h, and a mean prediction time of 36.8 minutes. Moreover, in the cases of focal seizures, the proposed model estimates the seizure focus.

The proposed model achieved a high capability in seizure prediction. Moreover, by using the automated feature selection of the deep learning algorithm, the patterns of the pre-ictal period in EEG signals were determined. Furthermore, by specifying the seizure focus, the model can help neurologists to take further curative actions.