Seizure Onset Zone Localization Method based on GNN Explanation

The localization of seizure onset zones from long term electrophysiological data is essential of epilepsy surgery, which demands significant effort and clinical knowledge to analysis long term electrophysiological data. In this study we proposed a Graph Neural Network (GNN)-based technique for automated localization of seizure onset zones in prolonged electrophysiological data. Initially, we established a Graph Convolutional Networks (GCN) model for epileptic stereo electroencephalography (SEEG) data classification. Then we used a GNN Explanation method to extract the magnitude of classification contribution from electrodes in the epilepsy SEEG data. The electrodes with most contribution would be considered as part of the seizure onset zone. The results revealed that the proposed method successfully identifies the primary seizure onset zones and estimates the impact of seizure foci on the other zones. This method applies graph neural networks explanation and offers novel perspectives for understanding the mechanisms of epilepsy.