Multi‐Metric EEG Connectivity and Graph Analysis Reveal Altered Resting‐State Networks in Major Depressive Disorder
Sushmit Jahan Rose, A. H. M. Rahmatullah Imon, Rushdania Bushra, Nafis Ahmed, Moha. Tasneem Habib WaridAbstract
Purpose
EEG functional‐connectivity studies in major depressive disorder (MDD) have produced inconsistent findings, partly because different estimators capture distinct aspects of neural coupling. We implemented a multi‐metric resting‐state EEG approach combining linear, phase‐based, nonlinear, and graph‐theoretical measures to characterize frequency‐dependent sensor‐level connectivity alterations and identify sensor‐level network patterns relevant to depression‐related dysconnectivity.
Method
Resting‐state EEG from 30 DSM‐IV MDD participants and 28 healthy controls was analyzed using a 19‐channel 10–20 montage. After MNE‐Python preprocessing, including filtering, ICA‐based artifact removal, current source density (CSD) transformation, and symmetric orthogonalization, coherence, PLV, PLI, and MI were computed across theta, alpha, and beta bands. Edge‐wise differences were tested using Brunner–Munzel with Holm correction, confirmed with Westfall–Young maxT correction, and summarized using Hedges’ g and graph measures.
Finding
MDD showed frequency‐ and metric‐dependent sensor‐level connectivity alterations. Theta coherence, PLV, and PLI were dominated by lower connectivity across frontal‐to‐posterior, fronto‐central, temporal/posterior‐temporal, parieto‐occipital, and central‐to‐occipital sensor pairs, whereas theta MI showed few increased frontal/fronto‐temporal effects. Alpha findings were more selective, with mainly lower coherence and PLV across frontopolar/frontal‐to‐central, fronto‐midline, and posterior links, while alpha MI showed a limited mixed pattern. In beta, coherence, PLV, and PLI showed sparse robust reductions, whereas beta MI showed widespread increased statistical dependence. Graph‐theoretical summaries at 20% proportional density showed mainly lower mean weighted strength and global efficiency in theta‐ and alpha‐band networks.
Conclusion
These findings indicate frequency‐ and metric‐dependent resting‐state EEG sensor‐level connectivity alterations in MDD. The dominant pattern was lower theta/alpha connectivity across distributed sensor pairs, alongside increased beta MI‐based dependence. The results provide corrected, effect‐size‐informed evidence of altered inter‐sensor coordination without establishing dysfunction of specific cortical regions or named networks. This framework may support future hypothesis‐driven research in independent, clinically annotated, higher‐density EEG or multimodal datasets.