An End-To-End Seizure Prediction Method Using Convolutional Neural Network and Transformer

With the fast development of intelligent medical technology, epileptic seizure prediction (SP) based on electroencephalography (EEG) has gradually become a frontier research topic in the field of healthcare digitalization due to the advantages of unravelling the mechanism of seizures and avoiding possible injuries. Existing SP methods based on EEG have several shortcomings such as poor feature representation ability and low prediction performance. In this paper, an end-to-end model for SP based on EEG is proposed, where convolutional neural network (CNN) is used to extract spatial features and transformer is employed to analyze long-term temporal information. The model proposed in this paper was evaluated on the Children's Hospital Boston-Massachusetts Institute of Technology (CHB-MIT) dataset. Experiment results show that the hybrid model in this paper achieves excellent seizure prediction performance, with accuracy of 92.5%, sensitivity of 91.8%, specificity of 93.1%, F1 score of 0.925 and area under the curve of 0.924.