Classification of recurrent major depressive disorder using a new time series feature extraction method through...

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Major Depressive Disorder (MDD) frequently recurs, necessitating early identification for effective prevention strategies.
• Recurrent MDD significantly impacts individuals, highlighting the need for improved diagnostic and prognostic tools.

Purpose of the Study:

• To develop and validate a novel machine learning approach for classifying recurrent Major Depressive Disorder (MDD).
• To identify brain regions and dynamic functional connectivity patterns associated with recurrent MDD using resting-state fMRI.

Main Methods:

• Extracted dynamic temporal features from resting-state fMRI data using an atlas-based approach.
• Utilized a Graph Convolutional Network (GCN) with time-series functional connectivity as adjacency matrices and brain region activity as node features.
• Employed Gradient-weighted Class Activation Mapping (Grad-CAM) to identify key brain regions contributing to the classification.

Main Results:

• Achieved a classification accuracy of 75.8% for recurrent MDD on the multi-site Rest-meta-MDD dataset.
• Identified specific brain regions implicated in the pathophysiology of recurrent MDD.
• Demonstrated the model's ability to capture dynamic changes in brain activity patterns.

Conclusions:

• The proposed GCN-based method effectively classifies recurrent MDD by integrating dynamic functional connectivity information.
• The study successfully identified brain regions crucial for distinguishing recurrent MDD, offering insights into its neural underpinnings.
• Further research into data pre-processing and harmonization may enhance classification performance.