Abstracts

Rationale: Intracranially-recorded high-frequency oscillation (HFO) during interictal slow-wave sleep is suggested to be a promising biomarker to localize the epileptogenic zone, partly because of the frequent occurrence in the seizure onset zone. A number of investigators developed automatic detectors to quantify the occurrence rate of HFO at given channels, and four are incorporated in RIPPLELAB for public use (Navarrete et al., 2016). We investigated the utility of HFO using the data from a cohort of 123 patients who underwent cortical resection following extraoperative ECoG recording. We determined whether each of the four different HFO detectors would improve the outcome prediction model solely based on the conventional clinical, seizure-onset zone (SOZ), and neuroimaging variables. Methods: The HFO rate at given channels was quantified by 1) short-time energy (STE) method (Staba et al., 2002), 2) short-line length (SLL) method (Gardner et al., 2007), 3) Hilbert method (Crepon et al., 2010), and 4) MNI method (Zelmann et al., 2012). In each patient, we calculated ‘subtraction HFO’ defined as the subtraction of HFO rate averaged across all preserved sites from HFO rate averaged across all resected sites. We expected that larger ‘subtraction HFO’ would be associated with a better outcome. We assessed the accuracy of prediction of patients achieving ILAE Class-1 outcome based on the full logistic regression model incorporating the clinical, SOZ, neuroimaging, and ‘subtraction HFO’ and that based on the reduced model incorporating all variables other than ‘subtraction HFO’. Results: Ninety patients had Class-1 outcome. ‘Subtraction HFO>80 Hz’ defined by each of the four detectors improved the performance of outcome prediction (STE: p=0.026; SLL: p=0.041; Hilbert: p=0.010; MNI: p=0.023). Neither ‘subtraction HFO>150 Hz’ nor ‘subtraction HFO>250 Hz’ improved the performance of outcome prediction (p>0.05). The area under the curve (AUC) on ROC analysis for prediction of Class-1 outcome was 0.767 in the reduced model. Incorporation of ‘subtraction HFO>80 Hz’ defined by STE, SLL, Hilbert, and MNI detectors improved the AUC to 0.793, 0.770, 0.797, and 0.807, respectively. Conclusions: The occurrence rate of HFO>80 Hz quantified by four different detectors equally and modestly improved the performance of prediction of postoperative seizure outcome in a cohort of 123 patients with focal epilepsy. Funding: NIH grant NS064033.