てんかん欠神における頭皮高頻度振動の表現型分類：K平均法クラスタリングを用いた複数の特徴量に基づく

Scalp high-frequency oscillations (HFOs) are promising noninvasive biomarkers of epileptogenicity, but their phenotypic diversity and clinical relevance in absence epilepsy (AE) remain unclear. This study aimed to classify scalp HFOs in AE using k-means clustering based on multiple morphological characteristics, and to evaluate their distribution across electroencephalogram (EEG) epochs and seizure control statuses. We analyzed scalp EEG recordings from 14 children and adolescents with AE. After excluding outliers, 163 scalp HFOs were characterized by average frequency, duration, amplitude, and number of cycles. Amplitude and cycle count were log-transformed prior to clustering, and k-means clustering was applied to identify distinct HFO phenotypes. Three clusters were identified: Cluster 1 (short duration, low amplitude), Cluster 2 (low frequency), and Cluster 3 (long duration, high cycle count). Cluster 2 and Cluster 3 were significant predictors of ictal HFOs in active AE, with odds ratios (ORs) of 0.33 (95% confidence interval [CI]: 0.14-0.74) and 5.00 (CI: 2.02-17.73), respectively. Cluster 2 also predicted interictal HFOs in active AE (OR [95% CI] = 2.71 [1.23-5.67]). These findings support the utility of scalp HFO phenotypes as EEG-based biomarkers for seizure detection and disease monitoring, potentially guiding treatment strategies in pediatric AE.