Related Experiment Video
Updated: Jun 3, 2025
08:20
MRI-guided dmPFC-rTMS as a Treatment for Treatment-resistant Major Depressive Disorder
Published on: August 11, 2015
13.8K
Predicting Antidepressant Treatment Response From Cortical Structure on MRI: A Mega-Analysis From the ENIGMA-MDD
Maarten G Poirot1,2,3, Daphne E Boucherie1,3, Matthan W A Caan2,4
1Amsterdam UMC, Department of Radiology and Nuclear Medicine, University of Amsterdam, Amsterdam, the Netherlands.
Human Brain Mapping
|January 6, 2025
Summary
Predicting antidepressant treatment response in major depressive disorder (MDD) using brain imaging alone was not successful. However, machine learning models showed promise in identifying response patterns in specific patient subgroups, particularly extreme responders or non-responders.
Area of Science:
- Neuroimaging
- Computational Psychiatry
- Pharmacogenomics
Background:
- Major depressive disorder (MDD) treatment response prediction is challenging, often involving lengthy trial-and-error.
- Individualized treatment selection could significantly improve patient outcomes and reduce healthcare costs.
- Cortical morphometry derived from structural MRI offers potential biomarkers for treatment response.
Purpose of the Study:
- To evaluate machine learning models for predicting individual antidepressant treatment response using cortical morphometry.
- To compare various machine learning pipeline configurations for predictive accuracy.
- To explore predictive performance in specific patient subpopulations and with the inclusion of subcortical data.
Main Methods:
- Utilized pooled longitudinal data from the ENIGMA-MDD consortium (n=262).
- Employed machine learning classifiers (gradient boosting, SVM, neural networks) with diverse data representations (regional averages, voxel-wise, projections).
- Investigated various cross-validation strategies (k-fold, leave-one-site-out) and inclusion of clinical/subcortical data.
Main Results:
- Overall prediction accuracy for antidepressant response was not significantly better than chance (balanced accuracy 50.5%).
- No machine learning pipeline configuration demonstrated superior performance.
- Exploratory analysis showed significant prediction (balanced accuracy 63.9%) in the extreme (non-)responders subpopulation.
Conclusions:
- Cortical structural MRI data alone is insufficient for reliably predicting individual antidepressant treatment response in MDD.
- Predictive models showed potential in identifying response patterns within specific MDD subpopulations.
- Future research should explore multimodal data integration for improved treatment response prediction.
