Localizing the Seizure Onset Zone with Bayesian Learning During iEEG Monitoring

A critical step in treating epilepsy patients is identifying the region where seizure activity originates: the seizure onset zone (SOZ). We present a novel approach to localizing the SOZ during ictal periods, i.e. seizures, in patients with drug-resistant epilepsy (DRE). We hypothesize that channels in the SOZ become prominent "sinks" in the brain network during seizures, as other regions attempt to inhibit the SOZ to terminate the seizure. Our proposed SOZ localization algorithm utilizes the sink index-a metric of how sink-like a channel is that is derived from dynamic network modeling- to iteratively update the probability of each channel being in the SOZ based on Bayes' theorem. We validated our approach using seizure events captured from two patients undergoing long-term intracranial EEG monitoring at Johns Hopkins Medical Center. Our Bayesian approach enhances the localization of SOZ by mathematically aggregating the sink index biomarker of the SOZ across seizures with various seizure onset and propagation patterns observed in DRE patients over multiple days of monitoring.