Abstracts

Rationale: High frequency oscillations are a promising biomarker of epileptic tissue, potentially providing additional information to guide resective surgery and improve outcomes. However, little is known regarding the spatiotemporal stability of the rate of these oscillations. We sought to quantify this variability in a large cohort of patients. Methods: We utilized a previously verified detector of automated high frequency oscillations that rejects artifacts and analyzed intracranial EEGs in 113 subjects: 91 patients with refractory epilepsy (interictal data, one night of 1-3 AM), 10 additional subjects with refractory epilepsy where longer recordings and full sleep scorings were available, and 12 control subjects without epilepsy (one night of 1-3 AM). Patients were classified as either having a) a single location of high HFO rate that was consistent over time; b) a single location that seemed to turned on and off over time; c) multiple, independent regions with high HFO rates that varied in dominance over time; and d) not enough HFOs present to make a determination. To quantify these observations, we developed an algorithm to automatically label each patient's data as belonging to one of these categories. This required quantification of the spatiotemporal distribution of HFOs, for which we developed a blind source separation algorithm using non-negative matrix factorization, which simultaneously determines groupings of channels and the temporal evolution of each group. Results: All four categories were found to be quite likely, with no category having a statistically significant difference in occurrence rate (0.05 threshold, ?2 test). In all patient cohorts, even those limited to interictal non-REM sleep and in patients without epilepsy, about 50% of patients were found to be in case (c)?"multiple regions with high rates of oscillations. In the patients with epilepsy, the variations were not always localized to the ictal onset zone or resected volume. Additionally, in the 10 patients with sleep scores, we scanned over the amount of recording time provided for the categorization algorithm, and observed that no amount of time could be definitively selected as sufficient?"even after a week of recording, newly recorded data could potentially change the number of observed sources of high frequency oscillations. Conclusions: Significant variation in HFO rates is present. Further research is needed to determine the influence of clinical and neural factors on these variations, as well as regarding how to best utilize rates of high frequency oscillations for clinical applications, given such variability. Categorizing HFOs simply by finding the regions of highest interictal rate may not be reliable for identifying the seizure onset zone. Funding: NINDS K08-NS069783, R01-NS094399, and Doris Duke Charitable Foundation to W.S. NIH K01-ES-026839 to S.G.