Data And Model Driven Localization Of The Epileptogenic Zone In Focal Epilepsy By Stereo Eeg

Epilepsy is a common chronic disease of central nervous system dysfunction.For patients with focal epilepsy whose seizures cannot be controlled by medication,surgical removal of the epileptogenic zone is one of the clinical cures for epilepsy.Using neuroimaging techniques to localize the epileptogenic zone is a necessity for the surgery to proceed,and stereotactic EEG as an invasive technique that can record deep intracranial electrophysiological signals can better assist in the preoperative evaluation when noninvasive findings are inconsistent.At present,clinical diagnosis mainly relies on the visual inspection by specialized physicians.The main purpose of this study is to automatically localize the epileptogenic zone in patients with focal epilepsy through stereotactic EEG signals,thus improving the diagnostic efficiency to a certain extent.Single-channel EEG signal features were combined with epileptic brain network features as input to the model.To extract epileptic brain network features,the epileptic brain network of a single patient was first constructed,obtained by setting the threshold value with the signal channel as the node and the phase locking value indicating the synchronization relationship as the edge;the average weighted degree,betweenness centrality and clustering coefficient of the interictal,pre-ictal and ictal brain networks of each patient were calculated as the epileptic brain network features.The frequency band energy obtained from wavelet packet decomposition with peak factor and pulse factor of interictal,pre-ictal and ictal periods were used as single-channel EEG signal features.Due to the data imbalance between the epileptogenic and non-epileptogenic channel samples,the model parameters were selected using hierarchical cross-validation,and a gaussian kernel support vector machine epileptogenic zone localization model was established,with an accuracy of 88.25%.To explore the possible causes of model misclassification,the EEG signal channels were considered in a larger intracranial context,and an interictal universal intracranial network framework was constructed to analyze the network dynamics of patients.The universal network framework was achieved by extracting the nodes and edges with high betweenness centrality that played a pivotal role in the interictal combined network of all patients.A corresponding patient whose channel was misclassified as an epileptogenic channel by the epileptogenic zone localization model was selected.The dynamics of the interictal original network and the new network combined with the universal network framework in this patient were analyzed by neural mass model,respectively.The epileptogenic EEG signal was simulated in the patient’s real seizure initiation channel,and the rest of the channels simulated normal EEG signal as the initial value to observe the propagation of the epileptogenic signal in the network.It was found that the EEG signal would be closer to the real situation in a larger range of networks.Improved epileptogenic zone localization model by obtaining new epileptogenic brain network features from interictal,pre-ictal and ictal combined networks respectively,the accuracy was improved to 90.98%,the precision was 80.00%,the sensitivity was 63.49% and the specificity was 96.70%.As a result,the model performance was improved from every aspect.The epileptogenic brain network and epileptogenic zone localization model proposed in this study can be used as an aid to clinical diagnosis of epilepsy and provide new insight into the preoperative evaluation of epilepsy.