Localizing epileptogenic network from SEEG using non-linear correlation, mutual information and graph theory analysis

Abstract

The key challenge in epilepsy surgery is precise localization and removal of the epileptogenic zone (EZ) from the brain. Localization of the epileptogenic network by visual analysis of intracranial EEG is extremely difficult. In this retrospective study, we used interictal connectivity and graph theory analysis on intracranial EEG to better delineate the epileptogenic zone. Patients who underwent surgery for drug-refractory mesial temporal and neocortical epilepsy were included. Computational measures, such as h2 nonlinear correlation and mutual information, were used to estimate the interdependency of intracranial EEGs. We observed that the Out-Degree, Out-Strength, and Betweenness centrality (graph properties) were the best predictors of EZ. From the results, we also found that graph properties with a normalized value above 0.75 were found to be a useful measure to localize the EZ with a sensitivity of 87.88 and a specificity of 87.13. Our results also validate that frequently occurring types of interictal fast discharges (IFD) with connectivity measures and graph properties can better localize the EZ. We foresee graph theory analysis of interictal intracranial EEG data can help precise localization of EZ for cortical resection as well as in minimally invasive radiofrequency ablation of epileptogenic hubs. Further, prospective validation is required for clinical use.