Identifying major depressive disorder based on cerebral blood flow and brain structure: An explainable multimodal...

Area of Science:

• Neuroimaging and computational psychiatry.
• Application of advanced machine learning techniques to neurological and psychiatric disorders.

Background:

• Magnetic resonance imaging (MRI) is crucial for non-invasive brain disorder assessment.
• Integrating multimodal MRI data enhances the detection of brain disorders.
• Major Depressive Disorder (MDD) diagnosis can benefit from advanced neuroimaging analysis.

Purpose of the Study:

• To identify Major Depressive Disorder (MDD) using a fusion of arterial spin labeling (ASL) perfusion MRI and structural MRI.
• To develop and validate an explainable fusion method for improved MDD detection.
• To analyze feature importance and interactions within the multimodal fusion model.

Main Methods:

• Collected ASL and structural MRI data from 260 participants (169 MDD patients, 91 healthy controls).
• Developed an explainable fusion model integrating cerebral blood flow (CBF) from ASL and brain tissue volumes from structural MRI.
• Analyzed feature importance and interactions to interpret the fusion model's decision-making process.

Main Results:

• The multimodal fusion model demonstrated superior predictive performance for MDD compared to individual modalities.
• Identified fourteen crucial features for MDD identification: eight regional volumes and six regional CBF measures.
• Discovered significant feature interactions, including three among important features and seven between structural and functional features.

Conclusions:

• Fusion learning integrating ASL and structural MRI data is effective for detecting MDD.
• The explainable AI approach reveals key features and their interactions, enhancing understanding of MDD pathophysiology.
• This multimodal approach offers a promising avenue for improving the diagnostic accuracy of MDD.