Abstracts

Rationale: High Frequency Oscillations (HFO) are a potential biomarker for epileptogenic tissue. However, HFO can also be present between ictal events or generated by normal tissue. Definitions of HFO for detection purposes are variable among investigators and characteristics of clinically relevant HFO are still to be defined. We propose a method to automatically detect HFO, their rate, and networks of co-occurrence to assess the seizure onset zone (SOZ) of focal seizures. Methods: Intracranial recordings from grid and depth electrodes from 7 consecutive patients undergoing presurgical evaluation were analyzed. For HFO detection we used an automated method that consists of identifying at least 4 consecutive oscillations with amplitudes greater than 10μV and at least two times larger than the average amplitude of oscillations in the surrounding background. Event duration threshold was 100ms to distinguish singular events from continuous runs of high frequency activity. The analysis was performed on bipolar EEG recordings, filtered at 80-200Hz for ripples and at 200-400Hz frequency range for fast ripples. We assessed HFO networks during interictal periods (> 1 hour before seizure onset), preictal (considered as 2 minutes prior to seizure onset) and ictal periods. We tested the significant co-occurrence of HFO events across channels, and their possible directionality using delay or other propagation measures. We compared results of these methods with SOZ channels indicated by clinicians by visual inspection of EEG and with SOZ assessment using HFO rate only. Results: Automatically detected HFO were used to rank channels by their HFO rates in each period, from the channels with the highest rates to the lowest to assess where SOZ channels were located. For each patient, time period, and method we constructed diagrams of significant co-occurrence and directionality. Networks of HFO co-occurrence usually included more channels than the SOZ (1-2 or 7 channels per patient only). During interictal periods, networks of HFO co-occurrence and channel ranking provided similar results: identification of the SOZ was accurate for 2 of the patients, and SOZ channels were within a greater network of HFO co-occurrence for 3 other patients. Compared to interictal periods, preictal analysis of co-occurrence of HFO were more consistent with the SOZ clinical assessment, and HFO networks were better matched to the SOZ (3 patients) than when only analyzing HFO rate (1 patient). During ictal periods with often fast propagation patterns, SOZ were correctly identified by both methods for 3 patients. However in 3 other patients, the highest HFO rate incorrectly identified the SOZ. For the same ictal periods, networks of HFO co-occurrence only misidentified SOZ in 1 patient. Conclusions: HFO seem to be a promising biomarker of the seizure onset zone, even when using interictal data and HFO rate analysis. However, the accuracy of SOZ identification can be increased when measuring the network of HFO co-occurrence during the periods leading to the onset of seizures and during seizures.