Rationale:
The human brain is increasingly understood as a complex network, and connectivity analysis offers a powerful framework to study drug-resistant epilepsy (DRE). While both invasive and non-invasive methods have examined network disruptions in epilepsy, most focus on low-frequency activity (< 80 Hz).
The contribution of high-frequency oscillations (HFOs) to brain network organization and their role in epileptogenesis remain largely unexplored. HFOs are promising biomarkers of the epileptogenic zone (EZ), yet their integration into network models—especially using non-invasive recordings—has received limited attention. Here, we apply Granger causality and graph theory to high-density EEG (HD-EEG) from children with DRE and typically developing controls (TDC) to map HFO-related effective connectivity and identify network features that may serve as non-invasive markers of the EZ.
Methods:
We analyzed HD-EEG (256-channels) data from 40 TDC (11 ± 3.2 y; 15 males) and 47 children with focal or generalized/diffuse DRE (12.4 ± 3.5 y; 17 males). Visually validated HFOs epochs were source localized, and activity was reconstructed across 204 regions of interest (ROIs), derived by subdividing 68 Desikan-Killiany atlas regions into three subregions (Fig.1A). For each HFO, we extracted the maximum activation time series per ROI and computed pairwise Granger causality to form 204×204 GC matrices. Adjacency matrices from GC results were used to construct Minimum Spanning Trees (MSTs) and node-level centrality metrics (Fig. 1B) were computed. Based on each HFO source localization maxima, events were grouped into: TDC (HFOs-TDC), epileptogenic regions (HFOs-EpiR), non-epileptogenic regions (HFOs-non-EpiR), and diffuse/generalized DRE (HFOs-Gen, Fig.1C). Results:
HFOs-EpiR showed consistently elevated network centrality across multiple metrics (Fig. 2). HFOs-EpiR showed significantly higher in-degree than HFOs-TDC (0.0021 vs 9.41e-4, p< 0.001), HFOs-non-EpiR (0.0021 vs 0.0016, p< 0.001), and HFOs-Gen (0.0021 vs 0.0013, p < 0.001); similarly, out-degree was higher in HFOs-EpiR compared to HFOs-TDC (0.0021 vs 9.41e-4, p < 0.001) and HFOs-non-EpiR (0.0021 vs 0.0016, p< 0.001). PageRank values were also elevated in HFOs-EpiR relative to TDC (0.0039 vs 0.0035, p< 0.001) and to HFOs-non-EpiR (0.0039 vs 0.0036, p< 0.001). Moreover, ground maximum was 0.40 for HFOs-EpiR vs 0.31 for HFOs-TDC, 0.35 for HFOs-non-EpiR, and 0.33 for HFOs-Gen (p < 0.001); ground mean was 0.15 for HFOs-EpiR vs 0.10 for HFOs-TDC, 0.11 for HFOs-non-EpiR, and 0.12 for HFOs-Gen (p< 0.001); and ground median was 0.14 for HFOs-EpiR vs 0.09 for HFOs-TDC, 0.097 for HFOs-non-EpiR, and 0.11 for HFOs-Gen (p< 0.001).