Localization of seizure onset zone with epilepsy propagation networks based on graph convolutional network

• An epilepsy propagation network was proposed to localize the SOZ. • Combine HFO propagation network and seizure propagation network. • A semi-supervised GCN based on pseudo-labelling and confidence was proposed. • 90% of the clinically diagnosed SOZ was covered using our proposed method. • Performance of our method is better than other quantitative SOZ localization methods. For the surgical treatment of epilepsy, localization of the seizure onset zone is essential. Studies have shown that the origin of epilepsy is closely related to the epileptogenic network in brain, so the epileptic activity propagation network may provide a new biomarker for localizing the seizure onset zone. In order to investigate the significance of epilepsy propagation networks for the localization and resection of seizure onset zone, a combination of high-frequency oscillation propagation networks and seizure propagation networks was processed through unsupervised and semi-supervised graph convolutional networks for comprehensive epilepsy propagation networks. The confidence coefficient of each electrode being located in the seizure onset zone was obtained, and electrodes closely related to the pathogenic epileptogenic network were located. The brain region localized using our proposed method covers 90% of the clinically diagnosed seizure onset zone. Compared to seizure propagation networks or high-frequency oscillatory propagation networks alone, the combined epilepsy propagation network based on graph convolutional networks was more strongly related with the possible epileptogenic brain region represented by the clinically diagnosed seizure onset zone. The epilepsy propagation network combined high-frequency oscillation propagation networks and seizure propagation networks may provide a better reference for prognosis.