Abstracts

Rationale: Brief (<100 ms), transient high-frequency waveforms (HFW) with characteristic frequencies >80 Hz have recently been estimated in non-invasive EEG from patients with focal epilepsy, including children. In contrast to high-frequency oscillations (HFO) in invasive signals, which have been specifically associated with the epileptogenic region and ictogenesis, the origin and role of non-invasively recorded HFW remain elusive. To be measurable at the scalp, they must originate from relatively large areas of the brain but it is unclear whether they are spatially distributed, related to seizure evolution, or represent a signature of the epileptic brain irrespective of seizure occurrence. Methods: Fifty nine pediatric patients (29 females), age 3-22 years [median = 9.5 years, (25^th, 75^th) quartiles = (6.0, 13.0) years] were studied. Fifty six had been diagnosed with medically refractory focal epilepsy and 3 had continuous recordings to evaluate spells, and based on normal results were included as controls. Scalp EEGs from epilepsy patients and controls spanned 43.2 – 161.1 h (median = 91.1 h) and 40.9 - 43.5 h (median = 42.7 h), respectively. All data were collected at Boston Children’s Hospital using a clinical 10-20 EEG system and were sampled at 512 (n = 3) or 1024 (n = 56) samples/s. Data were high-pass filtered with a 3^rd order elliptical filter (80 Hz cutoff). A 50-ms analysis window was used across patients. Random (noise-related) HFWs were excluded from further analysis. The remaining signals in individual EEG channels were classified using multiple classifiers, including spectral and subspace clustering. All data were analyzed in the Harvard Medical School High-Performance Cluster using the software Matlab (Mathworks, Inc). Results: Non-random, HFWs were detected in ~ 2 to 20% of analyzed segments. In epilepsy patients, HFWs were consistently classified in 6-10 clusters across channels, with central and occipital channels containing a lower number of clusters. In controls, HFWs also occurred in <20% of segments and were consistently classified in =3 clusters. At least 1 cluster contained artifact-related signals. The other 1-2 clusters contained HFWs of potentially physiological origin. The dominant signal frequencies in these two clusters were =120 Hz. Similar frequencies were estimated in 2-3 clusters from epilepsy patients but frequencies in the remaining clusters were significantly higher (up to ~180 Hz; p<0.01). In <50% of patients, there was a significant increase in the HFW rate (as high as 200 waveforms per min), prior to and during seizures in a subset of EEG channels that partially covered what was thought to be the seizure focus. However, in some patients, seizures occurred during periods of low HFW rates (<50 per min). These results were independent of the classification method used. Conclusions: Temporally sparse HFWs are measurable both in epilepsy patients and those with no detectable epileptiform activity and overall normal scalp EEG. Those estimated from epilepsy patients are significantly more heterogeneous (classified in a higher number of clusters) and with higher dominant frequencies. At least some of these clusters are likely to be associated with signal abnormalities that are specific to the epileptic brain. A statistical association between HFW occurrence and seizure evolution were consistent in a subset of epilepsy patients, with a significant increase in occurrence hours prior to seizure onset. Although preliminary, these results are from a relatively large cohort of pediatric epilepsy patients with scalp recordings spanning multiple days and suggest that at least some types of transient high-frequency signals may represent specific signatures of the epileptic brain, independently of seizure occurrence. Funding: National Science Foundation, Grant # ACI 1649865