Resting-state Functional Connectivity for Major Depressive Disorder Using Electroencephalography: A Deep Learning Approach

This research aimed to use the Convolutional Neural Networks (CNN) to identify biomarkers in Children and Adolescents with Major Depressive Disorder (MDD) apart from age-matched healthy young individuals. To achieve this, resting-state eyes-closed electroencephalography (EEG) was pre-processed, and the analysis was performed based on frequency bands and Regions Of Interest (ROI). We computed several resting-state Functional Connectivity (rsFC) measurements using a Multi-Variate Auto-Regressive (MVAR) model. We analyzed the best-performing model to understand the model decision-making process and identify the learned bio-markers. Overall, results achieved an F1-Score of 0.790 and a Mathew Correlation Coefficient (MCC) of 0.745 in the rsFC measurement of full frequency Partial Directed Coherence (ffPDC). We found that Partial Directed Coherence (PDC) performed better than Coherence, Partial Coherence, and Directed Transfer Function (DTF). We also found that the full-frequency version of PDC and DTF outperformed PDC and DTF and their other variants.