Abstracts

Rationale: Fast ripples (FR, 250-500 Hz) are high frequency oscillations (HFOs) in EEG signals which have been proposed as a specific biomarker of epileptogenic cortex. HFOs have conventionally been characterized in epileptic patients via invasive intracranial recordings. There is accumulating evidence that HFOs in the gamma and ripple range (80-250 Hz) are identifiable on scalp EEG. We aim to determine whether FRs are detectable on scalp EEG and whether scalp FR represent a biomarker of epilepsy. As HFO identification is time-consuming and subjective, we developed an automatic detector to facilitate and standardize scalp FR identification in a human-supervised, semi-automated system. Methods: Pediatric patients with tuberous sclerosis complex (TSC) with epilepsy selected from two multicenter, observational TSC studies and control pediatric patients, who underwent seizure evaluation and were determined to have no known brain-based diagnoses such as epilepsy, were included in this study. FR were identified in a single, 10-minute epoch during sleep for each patient in 2 independent, clinically blinded review stages: human visual review and semi-automated visual review. The automatic detector used for the semi-automated review phase utilizes a convolutional neural network to classify FR from putative events identified by an RMS detector as channel events containing localized increases in power. The human reviewer then verifies or rejects putative FR annotated by the automatic detector. The same 10-minute was used for each review stage. All FR per subject from both review stages were pooled for statistical analyses. We tested for difference in the presence or absence in FR between the epilepsy and control groups. Congruence of FR location with MRI tuber lobar location was determined. Spatial overlap between FR and epileptiform EEG abnormalities was evaluated. The performance of the automatic detector was assessed using human visual review annotated FR as the gold standard. Results: We included 7 pediatric patients with TSC-associated epilepsy and 4 normal controls. Seven out of 7 children with epilepsy had scalp FR compared to 0 out of 4 children in the control group (p=0.003, Fisher’s exact test). 86.5% [CI 74.7-93.3%] of FR occurred in lobar regions containing cortical tubers on MRI brain. Across all FR events, 61.5% [CI 48.0-73.5%] of FR were associated with spikes and 78.8% [CI 68.0-87.7%] of FR had partial overlap with channels that contained epileptic spike, spike-wave discharge, or paroxysmal fast activity. The automatic detector has sensitivity of 0.98 and false positive rate with range 2.7 to 20 false positives across all channels per minute per subject. Conclusions: Interictal scalp EEG-recorded FR occurred more frequently in children with TSC-associated epilepsy than in controls without epilepsy. Interictal scalp FR are detectable on scalp EEG and are a biomarker of epilepsy. The presented automatic detector achieves high sensitivity of FR detection, however implementation into a semi-automated approach with expert verification of the results to reduce number of false positives is advised. Funding: none