Graph Convolutional Neural Network based Depression Detection using Brain Functional Connectivity Measures

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Brain network connectivity analysis is crucial for understanding mental health, particularly major depressive disorder (MDD).
• Electroencephalogram (EEG) data offers insights into brain connectivity alterations in MDD, but its complexity challenges traditional analysis.
• Extracting meaningful information from EEG for accurate MDD diagnosis using machine learning remains a significant hurdle.

Purpose of the Study:

• To propose and evaluate a novel graph convolutional neural network (GCNN) model for classifying major depressive disorder (MDD) using scalp electroencephalogram (EEG) data.
• To identify dynamic functional connectivity patterns and neurophysiological markers associated with MDD.
• To integrate domain knowledge of brain regions with data-driven functional relationships for improved EEG analysis.

Main Methods:

• A graph convolutional neural network (GCNN) was developed for MDD classification using spectral coherence from resting-state and task-based EEG recordings.
• A unique graph representation for EEG data was created, integrating brain region knowledge with functional relationships.
• Extensive evaluations were performed on two distinct scalp-EEG datasets.

Main Results:

• The proposed GCNN model achieved high performance in classifying MDD patients, with an Area Under the Receiver Operating Characteristic curve (AUROC) of 93% for dataset 1 and 71% for dataset 2.
• Consistent neurophysiological biomarkers for MDD were identified, including dominant activity in left prefrontal, left frontoparietal, and right prefrontal regions.
• Delta, theta, and alpha frequency bands were found to be significant contributors, while the lower beta band showed minimal influence.

Conclusions:

• The novel GCNN approach effectively classifies MDD using EEG data by analyzing dynamic functional connectivity.
• The study identified specific brain regions and frequency bands as reliable biomarkers for MDD.
• This method offers a promising avenue for objective diagnosis and understanding of major depressive disorder.