Abstracts

Rationale: High frequency oscillations (HFOs), when observed in the electrocorticogram (ECoG), are believed to demarcate epileptogenic cortex. However, imaging these markers requires invasive procedures and good spatial coverage. The ability to detect HFOs noninvasively from the electroencephalogram (EEG) would be of great benefit to the diagnostic markup process. However, few studies have reported success in this endeavor and concerns that these events could be obscured artifacts need to be addressed. In this study, we report on a simple algorithm for isolating HFOs in scalp recordings from patients with refractory epilepsy prior to invasive presurgical evaluation. Methods: We acquired and analyzed two-hour long interictal, preictal, ictal, and postictal surface EEG recordings sampled at 1 kHz from seven patients with epilepsy at Stanford Medical Center with informed consent and prior IRB approval. The recordings were made from twenty-one electrodes affixed to the scalp according to the international 10–20 system; all were used for analysis except the two preauricular leads. The signals were referenced to the average of C3 and C4. Each signal was divided into 3-second non-overlapping epochs for analysis and an algorithm similar in spirit to one widely used for HFO detection (Staba et al., J Neurophysiol. 2002) from ECoG was applied as an initial screen. Candidate detections were further subjected to an upper threshold on r.m.s. power to eliminate large amplitude motion artifacts, which are common on surface EEG. Finally, a statistical measure of phase clustering, the Rayleigh index (RI), was computed and thresholded to further sift the detections into groups of events with and without significant rhythmicity. The concordance of this grouping made by the three-stage algorithm with a visual assessment was quantified in terms of conventional measures of sensitivity and specificity. Results: After applying the first two stages of the algorithm, a superset of 160 events were automatically detected, each with at least four oscillations concentrated in the 100-200 Hz frequency range. Visual inspection of this candidate set determined that 63.7% of the 160 events were true HFOs. After applying the third stage of the algorithm to separate events based on RI, true HFOs were detected with a sensitivity, specificity, and positive prediction value of 76.3%, 78%, and 82.2%, respectively. In general, it was found that as the threshold on RI was increased, so did the spatial concentration of HFO activity. This finding reinforces the belief that the events thus isolated are not merely artifacts. Conclusions: The results show that automated detection of HFOs on the scalp is feasible in patients with epilepsy. The rhythmicity of candidate detections appears to be an important factor in determining the epileptogenic zone. This assertion will be verified in our ongoing analysis of the data by comparing with the physician’s diagnostic evaluation when it becomes available. Funding: Support for this project came from National Science Foundation grants 1539068 and 1430833. AA received scholarship support from the Higher Committee of Education in Iraq. RSF is supported by the Maslah Saul Chair, the Anderson, Chen, Horngren funds for epilepsy research.