Related Experiment Video
Updated: Feb 28, 2026
05:19
Author Spotlight: Therapeutic Benefit of Closed-Loop Deep Brain Stimulation in Depression Treatment
Published on: July 7, 2023
3.6K
Applying machine-learning and deep-learning to predict depression from brain MRI and identify depression-related
Jiayue-Clara Jiang1, Camille Brianceau2, Elise Delzant2
1Institute for Molecular Bioscience, The University of Queensland, St Lucia, QLD, Australia. j.jiang@uq.edu.au.
Translational Psychiatry
|February 25, 2026
Summary
Machine learning models using brain grey-matter structure show limited accuracy for predicting major depressive disorder (MDD). While promising, these brain-based predictors require further research for clinical use in diagnosing depression.
Area of Science:
- Neuroimaging
- Psychiatry
- Machine Learning
Background:
- The accuracy of brain-based predictors for major depressive disorder (MDD) remains a significant challenge in clinical neuroscience.
- Existing methods often struggle to achieve high diagnostic performance using structural MRI data.
Purpose of the Study:
- To develop and evaluate machine-learning and deep-learning models for predicting MDD using voxel-wise grey-matter structure from T1-weighted MRI.
- To assess the performance of these predictors in independent UK Biobank cohorts and a clinical replication cohort.
Main Methods:
- Trained Best Linear Unbiased Predictors (BLUP) and ResNet3D deep-learning models on UK Biobank data (987 MDD cases, 3934 controls).
- Evaluated predictor accuracy using Area Under the Curve (AUC) in an independent UK Biobank sub-cohort (483 MDD cases, 1939 controls) and the DEP-ARREST CLIN cohort (64 MDD cases, 32 controls).
- Compared BLUP predictor performance with a polygenic score (PGS) for major depression.
Main Results:
- The BLUP predictor showed a modest but significant association with MDD status in the UK Biobank (AUC=0.57, p=1.1×10⁻⁵).
- Analysis of brain regions of interest (ROIs) suggested contributions from areas like the hippocampus and amygdala, though not reaching multiple testing correction.
- The BLUP predictor captured unique variance not explained by the PGS, with a combined AUC of 0.66.
- The deep-learning predictor did not show significant association with MDD after multiple testing corrections.
- Estimated morphometricity of MDD was 0.061, indicating limited potential for grey-matter structure-based prediction (maximal AUC=0.64).
Conclusions:
- Grey-matter structure-based predictors for MDD currently have limited clinical utility due to modest accuracy.
- The BLUP approach shows potential, capturing genetic and non-genetic aspects of MDD, warranting further investigation.
- Future research should focus on integrating multimodal data and exploring brain-region contributions to improve MDD prediction models.
