Intracranial High-Frequency Oscillations and Epileptogenic Zone: Incorporating Neuroanatomic Variation

Purpose: To determine if incorporating neuroanatomic or intersubject variation in the occurrence rate of intracranial high-frequency oscillations improves its diagnostic performance for localization of epileptogenic zone (EZ). Methods: Five minutes of awake stereo-electroencephalography data from 59 patients were analyzed. High-frequency oscillations were analyzed using three different normalization methods: rate per minute, by neuroanatomic region across the patient population, and patient-wise. Generalized linear mixed effects models were trained in patients with good seizure outcomes after epilepsy surgery (higher confidence in the clinical localization of EZ) and tested in patients with poorer outcomes (validation set approach). Results: The generalized linear mixed model with region-wise normalization across the patient population best localized the EZ (highest area under the curve 0.69), closely followed by the rate per minute (0.68). In the test subgroup, the optimal generalized linear mixed model predicted EZ in individual patients with an accuracy of 0.18 to 0.86, sensitivity of 0.05 to 1.00, and specificity of 0.12 to 0.95. In patients with poorest performance of the generalized linear mixed model, although the electrode contacts within EZ were correctly identified, there was a high number of false positives (model-predicted electrode contacts lying outside clinically ascertained EZ). Model performance varied across neuroanatomic regions, with the highest accuracy i