Toward neuroimaging-based diagnostic support: a deep learning approach with a closed-loop system for psychiatric...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Psychiatric Diagnostics

Background:

• Neuroimaging provides insights into psychiatric disorders but faces challenges in clinical translation.
• Current diagnostic methods for psychiatric disorders can be subjective.

Purpose of the Study:

• To develop and validate a deep learning framework for classifying multiple psychiatric disorders using structural MRI.
• To assess the generalizability and real-world applicability of the developed model.
• To integrate the model into a clinical workflow for diagnostic support.

Main Methods:

• Development of the patch-based hierarchical network (PHN), a deep learning framework.
• Training the PHN on a large dataset (n=2,490) including four major psychiatric disorders and controls.
• Validation on independent research datasets (n=1,346) and real-world clinical data (n=344).

Main Results:

• The PHN demonstrated robust performance in classifying psychiatric disorders.
• The model showed generalizability across diverse datasets, including real-world clinical data.
• The system successfully reflected complex clinical presentations, such as comorbidities.

Conclusions:

• The patch-based hierarchical network (PHN) offers a promising approach for objective psychiatric diagnosis using structural MRI.
• Integration into a clinical workflow represents a significant step towards bridging the research-to-practice gap in psychiatric neuroimaging.
• AI-powered neuroimaging analysis can provide valuable objective support for clinicians in diagnosing psychiatric disorders.