EEG-based depression detection using optimized brain network via node betweenness centrality

Resting brain networks linked with aberrant functional connectivity in clinical depression have been assessed using fMRI. However, weak temporal resolution and high equipment cost do not support the diagnostic credibility. Alternatively, the flexible topology over specific spatiotemporal scales can be accessed using Electroencephalography (EEG} by projecting the co-activated brain regions of resting brain networks onto the scalp. This research proposes a novel approach to optimizing resting brain networks that effectively minimizes feature space's dimensionality to detect Major Depressive Disorder (MDD) accurately. EEG node reduction is based on a robust graph centrality algorithm, i.e., node betweenness centrality, to quantify the importance of nodes within the resting brain networks. The findings of this study suggest that the differential centrality index explicitly identified the most influential nodes responsible for classification along with 58.5% data reduction. Moreover, between-group differences were dominated by the beta (15-30 Hz) band dynamics, which uniquely discloses the neural correlates of MDD phenotypes.