EEG Connectivity Signatures in Migraine and Tension-Type Headache: A Multimethod ROI-Based Functional Connectivity and Machine-Learning Analysis
TL;DR
The results point to method-dependent, region-specific sensor-level EEG differences across migraine, TTH, and control groups that are best viewed as candidate neurophysiological signatures rather than clinically ready biomarkers.
Abstract
Background/Objectives: Migraine and tension-type headache (TTH) are common primary headache disorders, but their underlying network-level EEG patterns remain incompletely characterized. We examined whether resting-state functional connectivity measured with complementary estimators, predefined sensor-level regions of interest (ROIs), graph features, and machine-learning models could distinguish migraine, TTH, and healthy controls. Methods: The study included 150 participants (61 migraine, 47 TTH, and 42 controls). Connectivity was calculated in the delta, theta, alpha, beta, and gamma bands using coherence, imaginary coherence, weighted phase-lag index (wPLI), and debiased wPLI. We evaluated global, regional, topographic, and graph-theoretical features, performed ROC-AUC analyses, and tested classification models with nested cross-validation. To limit the multiple-testing burden, false discovery rate correction was applied to a prespecified, hypothesis-driven ROI set. Results: Eight candidate ROI features remained significant after correction. TTH showed the highest gamma temporo-parietal coherence, whereas migraine showed higher gamma temporo-parietal wPLI and debiased wPLI and a higher exploratory migraine probability-like score. Controls had higher delta fronto-temporal and gamma fronto-temporal/temporo-parietal imaginary coherence. Single-feature ROC-AUC values were moderate, and the nested machine-learning models showed only modest classification performance. Conclusions: The results point to method-dependent, region-specific sensor-level EEG differences across migraine, TTH, and control groups. They are best viewed as candidate neurophysiological signatures rather than clinically ready biomarkers. External validation, fuller clinical covariate assessment, and source-level analyses are needed before diagnostic use can be considered. Future studies should also examine whether these candidate signatures differ according to aura status and episodic or chronic headache subtype.