Automatic Seizure Prediction Based on Cross-Feature Fusion Stream Convolutional Neural Network

Seizure is a common nervous system disease, currently about 1% of the world’s population suffer from seizure. EEG signals are the main tools for predicting seizures. Methods to accurately predict seizures would help reduce helplessness and uncertainty. In this paper, we designed a convolutional neural networks (CNNs) based on cross-feature fusion stream for seizure prediction using seizure datasets from Boston Children’s Hospital. The EEG data collected in time domain, frequency domain and time frequency domain were fused with the algorithm to classify the preictal and interictal so as to predict seizure. Experimental results show that the cross-feature fusion stream CNN model achieves 97% accuracy on the CHB-MIT dataset.