Graph-Based Analysis of Brain Connectivity in Major Depressive Disorder Using EEG

In line with our prior research, we aimed to identify and classify biomarkers of major depressive disorder (MDD) using electroencephalography (EEG) in patients and age-matched healthy children and adolescents. To achieve this, resting-state eyes-closed EEG data were pre-processed, and the analysis was performed on a channel-wise and Regions of Interest (ROI)-wise basis. We computed several resting-state Functional Connectivity (rsFC) measurements using a Multivariate Auto-Regressive (MVAR) model. The resulting connectivity measurements were then used to train a Random Forest model. We analyzed the best-performing models to understand the model's decision-making process and identify the learned biomarkers. Overall, the Random Forest model achieved a F1-score of 0.975 and a Matthews Correlation Coefficient (MCC) of 0.970. The model's test scores were highest with Partial Directed Coherence (PDC) and its variants. Additionally, we found that direct measurements, specifically PDC Factor and direct Directed Transfer Function (dDTF), performed better than other variants.