Machine learning framework for depression subtype grouping: integrating high-resolution imaging and clinical symptom...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Major depressive disorder (MDD) affects a significant portion of the population, with many experiencing treatment resistance.
• Current depression classification relies on subjective symptom severity, limiting treatment prediction.
• There is a need for objective, data-driven methods to classify MDD subtypes.

Purpose of the Study:

• To develop and validate a novel machine learning (ML) framework for classifying MDD subtypes.
• To integrate clinical features with high-resolution magnetic resonance imaging (MRI)-derived features for unbiased classification.
• To identify distinct, potentially predictive MDD subtypes.

Main Methods:

• Utilized canonical correlation analysis (CCA) to identify correlations between clinical and MRI-derived features.
• Employed hierarchical clustering to group participants based on derived clinical-imaging phenotypes.
• Analyzed data from 64 MDD participants using 11 clinical assessments and 7T T1-weighted MRI scans.

Main Results:

• Identified three highly correlated clinical-imaging phenotypes: anhedonia-brainstem, childhood trauma-anhedonia-right frontal pole, and distress-right temporal lobe.
• Hierarchical clustering revealed two distinct MDD clusters: one with high childhood trauma scores and another with scores similar to healthy controls.
• The ML framework provided a data-driven, non-biased classification of MDD.

Conclusions:

• The study presents a novel ML framework integrating clinical and neuroimaging data for MDD classification.
• This approach offers a more objective and potentially predictive alternative to traditional symptom-based depression subtypes.
• The identified clusters may aid in understanding MDD heterogeneity and guiding treatment strategies.