Multi-Granularity Graph Convolution Network for Major Depressive Disorder Recognition.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Machine Learning

Background:

• Major Depressive Disorder (MDD) is a prevalent psychological condition.
• Existing machine learning methods for MDD recognition using EEG often filter out potentially relevant weak brain connections.
• Current statistical features fail to capture complex topological and propagation patterns in brain networks.

Purpose of the Study:

• To develop a novel method for improving Major Depressive Disorder (MDD) recognition accuracy using resting-state EEG signals.
• To address limitations of existing methods in handling weak functional brain connections and capturing network topology.
• To propose a Multi-Granularity Graph Convolution Network (MGGCN) for enhanced MDD detection.

Main Methods:

• Proposed a Multi-Granularity Graph Convolution Network (MGGCN) for MDD recognition.
• Constructed a multi-granularity functional neural network using multiple thresholds to preserve weak connections.
• Utilized graph neural networks to learn topological structure and brain saliency patterns from EEG data.

Main Results:

• The MGGCN method demonstrated superior performance and efficiency on benchmark datasets.
• Analysis revealed increasing connectivity defects in specific brain regions (RF, RT, LT, LP) with increasing granularity.
• Identified brain functional connections in these regions as potential biomarkers for MDD.

Conclusions:

• The MGGCN approach effectively retains valuable weak connections while mitigating noise for improved MDD recognition.
• The study highlights the potential of specific brain connectivity patterns as biomarkers for Major Depressive Disorder.
• MGGCN offers a promising tool for advancing the computational diagnosis of depression.