Epileptic Seizure Detection Based on Semi-supervised Generative Adversarial Network

Abstract Since the manual diagnosis of electroencephalograph (EEG) recordings requires a lot labor and material costs for clinical seizure detection, annotation data is great challenge detection. To tackle issue small samples epilepsy-labeled data, we propose semi-supervised generative adversarial network-based detection method. begin with, Butterworth filter used to preprocess EEG, filtered EEG signal input into SGAN model. Finally, output model subjected post-processing operations including averaging filtering threshold comparison, discriminative result whether tested output. The method achieved an average sensitivity 90.36%, specificity 93.72%, accuracy 93.72% in CHB-MIT dataset. Experiments demonstrate that network has more accurate classification performance epilepsy