Source-Level Multiscale Granger Causality from Preictal Scalp EEG Identifies Candidate Seizure-Onset Zones in Drug-Resistant Focal Epilepsy

Background/Objectives: Precise localization of seizure-onset zones (SOZs) from noninvasive electroencephalography (EEG) remains challenging in the presurgical evaluation of drug-resistant focal epilepsy. We investigated whether cortical source reconstruction combined with multiscale spectral Granger causality (GC) analysis of preictal scalp EEG could identify candidate SOZ regions. Methods: Sixteen participants underwent presurgical evaluation; two were excluded for excessive EEG artifacts, leaving 14 participants (seven resected with Engel class I outcomes and seven unresected). A clinically representative 20 s preictal EEG segment was cleaned using independent component analysis, EEG source solutions were reconstructed with sLORETA on a template head model, and the EEG source activities were then parcellated into 68 Desikan–Killiany cortical regions. Nonparametric GC using Janashia–Lagvilava spectral matrix factorization was estimated in four bands (1–12, 13–30, 30–45, and 55–95 Hz). Regional causal inflow, outflow, and net flow were calculated. Candidate SOZ regions were defined as those exceeding the 95th percentile of absolute net-flow values. Concordance was assessed against clinically determined SOZ localization and, in resected cases, postoperative surgical outcome. Results: Using beta (13–30 Hz) as the primary band, we found that candidate SOZ regions were concordant with clinical localization in 6/7 resected participants (85.7%; exact 95% CI, 42.1–99.6%) and 3/5 scorable prospective participants (60.0%; 14.7–94.7%), yielding 9/12 overall (75.0%; 42.8–94.5%). Two prospective participants lacked sufficient clinical localization for assessment. Three resected participants with left mesial temporal epilepsy showed reproducible left parahippocampal dominance across beta, gamma, and high-gamma bands, with variable entorhinal co-involvement. Top-ranked candidate SOZ regions consistently exhibited net causal inflow. Conclusions: In this small, heterogeneous cohort, source-level net-flow mapping of preictal scalp EEG identified clinically plausible candidate SOZ regions; these exploratory findings suggest a potential noninvasive adjunct to presurgical evaluation. The inflow-dominant pattern supports a receiver-like role of the SOZ. Larger prospective studies with individualized head models, multiple seizures, and intracranial EEG validation will help establish robustness and clinical utility.