Evolutionary neural architecture search for automated MDD diagnosis using multimodal MRI imaging

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Psychiatry

Background:

• Major depressive disorder (MDD) significantly impacts global health.
• Neuroimaging techniques like MRI offer potential for MDD diagnosis.
• Current computer-aided diagnosis relies heavily on expert experience, limiting scalability.

Purpose of the Study:

• To develop an automated framework for diagnosing MDD using multi-modal MRI.
• To overcome limitations of traditional computer-aided diagnosis methods.
• To identify key brain regions involved in MDD diagnosis.

Main Methods:

• Proposed M-ENAS (Multi-modal Evolutionary Neural Architecture Search) framework.
• Employed a two-stage search: one-shot NAS for supernet weights and evolutionary search for network architecture.
• Validated M-ENAS on two independent datasets.

Main Results:

• M-ENAS demonstrated superior performance compared to existing hand-designed diagnostic methods.
• The framework successfully automated the diagnosis of MDD using multi-modal MRI data.
• Identified specific brain regions within the somatomotor network crucial for MDD diagnosis.

Conclusions:

• M-ENAS provides an effective and automated approach for MDD diagnosis using neuroimaging.
• The findings offer insights into the neurobiological underpinnings of MDD.
• Automated architecture search can enhance the accuracy and efficiency of diagnostic tools.