Multi-modal MRI for objective diagnosis and outcome prediction in depression

Area of Science:

• Neuroimaging
• Biomarkers
• Machine Learning

Background:

• Major depressive disorder (MDD) diagnosis relies heavily on subjective clinical assessments, leading to low treatment effectiveness.
• Objective biomarkers from magnetic resonance imaging (MRI) could aid clinical decision-making in MDD.
• Existing MRI-based biomarkers are often uni-modal and focus on either diagnosis or outcome prediction, not both.

Purpose of the Study:

• To identify multi-modal MRI-based predictors for both MDD diagnosis and 6-month treatment outcome.
• To compare uni-modal and multi-modal classification approaches for MDD diagnosis and outcome prediction.
• To investigate the utility of baseline MRI features for predicting treatment response in MDD.

Main Methods:

• Acquired structural (T1-weighted, T2-weighted, DTI) and functional (resting-state fMRI) MRI scans from 32 MDD patients and 31 healthy controls (HC) at baseline.
• Extracted features from baseline MRI scans and defined 6-month outcome classes (negative vs. positive) based on depression severity changes.
• Employed Support Vector Machine (SVM) models for uni-modal and multi-modal classification of MDD vs. HC (diagnosis) and negative vs. positive outcome.

Main Results:

• Diffusion tensor imaging (DTI) features showed the highest uni-modal performance for diagnosis (mean diffusivity, AUC=0.701) and outcome prediction (sum of streamline weights, AUC=0.860).
• Multi-modal ensemble classifiers integrating T1-weighted, resting-state fMRI, and DTI features significantly improved classification performance for both diagnosis (AUC=0.746) and outcome (AUC=0.932).
• Key predictive features were localized in frontal, limbic, and parietal brain areas, with distinct modalities and locations for diagnostic versus prognostic models.

Conclusions:

• Combining MRI features from multiple modalities enhances the prediction accuracy for both MDD diagnosis and treatment outcome.
• The most influential features for MDD diagnosis differ in location and modality compared to those predicting treatment outcome.
• This study provides objective MRI-based biomarkers for MDD diagnosis and outcome prediction, warranting further validation in larger cohorts.