Authors :
Presenting Author: Zhengxiang Cai, PhD – Carnegie Mellon University
Colton Gonsisko, BS – Carnegie Mellon University
Xiyuan Jiang, PhD – Carnegie Mellon University
Anto Bagic, MD, PhD – University of Pittsburgh Department of Neurology
Gregory Worrel, MD,PhD – Mayo Clinics
Mark Richardson, MD, PhD – Massachusetts General Hospital
Bin He, PhD – Carnegie Mellon University
Rationale: Despite advances in epilepsy surgery, 40-50% of epilepsy patients fail to achieve seizure freedom due to incomplete epileptogenic zone (EZ) delineation. Current biomarkers—spike and HFO rates—show high variability and limited specificity. Even improved biomarkers, such as spike-HFO coupling, measure only occurrence without considering the spatiotemporal propagation dynamics of epileptic networks. Therefore, we hypothesized that millisecond-resolved HFO sequences—rapid propagating chains across cortical regions—would reveal functional network organization, enabling superior EZ localization and outcome prediction.
Methods:
We analyzed 101,186 channel-hours of intracranial EEG from 40 patients (mean 23 h) with drug-resistant focal epilepsy across two epilepsy centers (Mayo Clinic, University of Pittsburgh). Using an automated detection algorithm with unsupervised clustering, we identified over 8.9 million putative HFOs and sequenced them based on spatiotemporal patterns, yielding 273,697 HFO sequences with an average of 7.73 events per sequence spanning 67.58 ms. We compared HFO sequence mapping against conventional HFOs, interictal spikes, and coupled spike-HFOs for clinical EZ localization and outcome prediction. Directional connectivity analysis using directed transfer function quantified information flow during HFO propagation.
Results:
HFO sequences (HFO-seq) demonstrated superior performance across all metrics. For EZ localization, HFO-seq achieved 2.92 mm error—an 818% improvement over conventional HFO rates (18.57 mm) and markedly better than both spike-coupled HFOs (10.11 mm) and spike rates alone (18.21 mm; all p< 10-3). While HFO sequences and spike-coupled HFOs showed similar F1 scores (0.64), HFO sequences exhibited higher sensitivity, capturing more of the true EZ compared to HFOs and spikes (both F1: 0.56). Importantly, HFO-seq mapping remained stable with just 30-minute recordings (SD: 3.23mm), enabling clinical feasibility. For surgical outcome prediction, HFO-seq achieved 0.90 AUC (84% accuracy, 95% precision), outperforming conventional HFO rates (0.72), spike rates (0.73), and spike-coupled HFOs (0.79). Connectivity analysis revealed predominant inward information flow from periphery to HFO zones, in seizure-free patients (information asymmetry: -0.21, p< 0.01) but not in non-seizure-free patients (-0.03, p=0.43), suggesting differential network suppression mechanisms underlie surgical outcomes.