An Optimized EEG-based Intrinsic Brain Network for Depression Detection using Differential Graph Centrality

Millions of adults suffer from Major Depressive Disorder (MDD) globally. Applying network theory to study functional brain dynamics often uses the fMRI modality to identify the perturbed connectivity in depressed individuals. However, the weak temporal resolution of fMRI limits its ability to access the fast dynamics of functional connectivity (FC). Therefore, electroencephalography (EEG), which can track functional brain dynamics every millisecond, may serve as a diagnostic marker for utilizing the dynamics of intrinsic brain networks at the sensor level. This research proposes a unique neural marker for depression detection by analyzing long-range functional neurodynamics between the default mode network (DMN) and visual network (VN) via optimal EEG nodes. While DMN abnormalities in depression are well documented, the interactions between the DMN and VN, which reflect visual imagery at rest, remain unclear. Subsequently, a novel differential graph centrality index is applied to reduce a high-dimensional feature space representing EEG temporal neurodynamics, which produced an optimized brain network for MDD detection. The proposed method achieves an exceptional classification performance with an average accuracy, f1 score, and MCC of 99.76\%, 0.998, and 0.9995 for the MODMA and 99.99\%, 0.999, and 0.9998 for the HUSM datasets, respectively. The findings of this study suggest that a significant decrease in connection density within the beta band (15-30 Hz) in depressed individuals exhibits disrupted long-range inter-network topology, which could serve as a reliable neural marker for depression detection and monitoring. Furthermore, weak FC links between the DMN and VN indicate disengagement between the DMN and VN, which signifies progressive cognitive decline, weak memory, and disrupted thinking at rest, often accompanied by MDD.