Abstracts

Rationale: Interictal high-frequency oscillations (HFO, 80-150 Hz) in scalp EEG are currently investigated as a noninvasive biomarker of epileptogenic activity but their diagnostic power in reviewing scalp EEG remains unclear, especially in children. In this study, we correlate the spatial distribution of HFO across channels with the occurrence of spikes in periods of interictal wakefulness and sleep in children with epilepsy. A computational tool for the unsupervised annotation of HFO in scalp EEG is developed. Methods: The study includes 33 children (age: 8.6±5.0 [mean±SD], 21 male, 12 female) with heterogeneous etiologies and epilepsy diagnoses who underwent scalp EEG monitored at the Connecticut Children Medical Center from December 2017 to March 2018. For each patient, one epoch of EEG during wakefulness and one epoch during sleep (stage N2-N3) were considered (wakefulness: 15.1±5.1min; sleep: 15.6±3.3min [mean±SD]; 10-20 configuration, bipolar montage; 1,024 Hz sampling rate). Each epoch was reviewed offline by a board-certified pediatric epileptologist who marked the presence of spikes by visual inspection independently of this study. The proposed tool was applied to each EEG epoch in two steps. First, a multi-objective HFO detector was applied to identify candidate events that show increments in amplitude discharge and high frequency activity compared to the background. Then, 7 time-frequency features were computed for each candidate and the resultant feature vectors were clustered according to the Ward’s method to remove artifacts (unsupervised clustering). The candidates that passed the clustering stage were putative HFO and their rate per minute was assessed. Results: A total of 4,377 candidate events were analyzed and 4,006 (91%; sleep: 783 out of 872; wakefulness: 3,223 out of 3,505) were marked as putative HFO at the end of the clustering procedure, which resulted in 0.2±0.4 ripples/min per channel. Spikes were identified in 19 out of 33 patients (asleep: 19; awake: 17), with 2.09±0.98 channels with spikes per patient (awake: 2.0±0.97; asleep: 2.18±1.01 [mean±SD]). In the asleep case, the spikes were detected in the channels with the highest HFO rates and the average HFO rate was higher in spike-bearing channels than spike-free channels in 13 out of 19 patients, with significance in 7 out of 13 (paired t-test, P-value P<0.03, Figure 1A). Across patients, the HFO rate was significantly higher in spike-bearing channels than spike-free channels during sleep (P<0.02), but not wakefulness (P=0.58), Figure 1B. Finally, no significant HFO rate difference was reported between spike-free channels across the entire patient population (Figure 1C). Conclusions: This is one of the first studies to characterize the temporal arrangement of scalp HFO compared to interictal spikes in children during wakefulness and sleep. A robust method for the unsupervised HFO annotation is validated and the results indicate that scalp HFO can be related to spikes during sleep, which is relevant for the EEG-informed diagnostic of pediatric epilepsy. Funding: IBACS Seed Grant 24, Connecticut Institute for the Brain and Cognitive Science, University of Connecticut.