Epilepsy seizure detection using Transformer

Abstract

Electroencephalogram (EEG) is a general examination method for doctors to diagnose epilepsy, and it is also an important tool for studying brain activity. Due to the time-consuming and uncertainty of manually extracting features from EEG signals, this paper will be based on an end-to-end deep learning method different from the classic CNN and RNN network structure. This paper uses a relatively novel Transformer network structure to identify EEG whether the signal is epileptic. The experiment in this paper was carried out on the public CHBMIT data set, and finally, the average result of the five-fold cross-validation was 94.46%, the specificity was 93.97%, and the sensitivity was 94.96%. The experimental results show that the Transformer model has a higher performance improvement than the classic Resnet and Bi-LSTM networks, and it has greater potential in future epilepsy detection applications.