Abstracts

Rationale: Resective surgery remains the main treatment option for approximately one quarter of patients with epilepsy who fail to obtain control of their seizures through pharmaceutical interventions. This procedure requires invasive surgery and long hospital stays while awaiting spontaneous seizures. One emerging method of potentially improving surgery outcomes and reducing the hospital time involves using high bandwidth EEG to identify additional biomarkers such as high frequency oscillations (HFOs). A growing collection of data indicate that interictal HFOs are correlated with the seizure onset zone. However, clinical use of HFOs remains very challenging for many reasons, including 1) the big-data challenge of efficiently processing the high resolution data, 2) distinguishing neural HFOs from non-neural artifacts, 3) determining prospective rules of determining which and how many electrodes have an "abnormal" amount of HFOs. Methods: Innovations directly addressing each of these challenges are presented, specifically 1) a new data processing paradigm and implementation based on procedures used in particle physics detectors, 2) novel methods to identify artifactual HFO detections, and 3) a general algorithm using Kernel Density Estimation to determine whether the HFO rates allow prediction of the seizure onset zone and if so, the specific prediction. The procedures are applied to data from 27 patients recorded at multiple centers. The amount of recording time needed for a prediction is estimated by using multiple random subsamples of the data of fixed time lengths. Results: The efficacy of this procedure is quantified via a) the false positive rate, b) the typical amount of recording time required to make a prediction, and c) the percentage of patients for which the algorithm is applicable. In nearly half of the patients, the blinded algorithm identified a subset of electrodes that are likely within the seizure onset zone with very high precision. In the remainder of patients, the algorithm determined that HFO rates were nondiagnostic. Conclusions: Based on this initial study of 27 patients, the new algorithm is very specific at determining the seizure onset zone and is applicable for a large portion of patients undergoing resective surgery.