Bridging computational and clinical strategies for presurgical identification of epileptogenic networks

About one third of epilepsy patients are drug-resistant. Resective epilepsy surgery remains a key treatment option but depends critically on accurate identification of the seizure onset zone (SOZ), which is still guided mainly by subjective visual inspection of electrophysiological signals. Network-based metrics derived from intracranial EEG have recently shown promise for SOZ identification, but their evaluation and interpretation have remained disconnected from standard clinical procedures and reasoning.

We analyzed stereotactic EEG (sEEG) recordings from 20 patients undergoing presurgical evaluation in the interictal state and during clinical mapping via electrical stimulation. We constructed patient-specific time-varying dynamic network models and addressed key questions for clinical translation: how sensitive network vulnerability estimates are to the choice among related published metrics, how they depend on the time of the day, and how the resulting conclusions relate to stimulation-evoked epileptiform discharges as a routine clinical reference for network vulnerability in 5 patients who underwent 50 Hz stimulation mapping. We then simulated virtual thermocoagulation in 6 patients who later underwent thermocoagulation by removing the clinically coagulated nodes and testing whether the resulting network changes went beyond pure network size reduction.

The network metrics correlated with epileptiform discharges evoked by 50 Hz intracranial stimulation in four of five stimulated patients, supporting a link between model-based network fragility and interictal epileptiform discharges evoked in clinical stimulation mapping. Using virtual thermocoagulation, we quantified the expected network-level change under model node removal, capturing both local and global effects depending on individual network architecture. Across patients, more fragile network metrics pointed toward clinically defined SOZ contacts and yielded stable conclusions across time, conditions, and perturbation properties, supporting their reliability.

Together, these findings provide a clinically interpretable calibration of published network vulnerability metrics against routine clinical references, using interictal sEEG data only.

Some people with epilepsy need brain recordings to find the tissue where seizures start. We tested whether computational models built from seizure-free sEEG recordings can identify vulnerable parts of the epileptic network. The model-based measures pointed toward clinically defined seizure onset regions, were stable across recording times, and often agreed with responses seen during clinical brain stimulation. These results suggest that interictal network modeling may complement standard presurgical evaluation, although larger prospective studies are needed.