Predicting depression in healthy young adults: A machine learning approach using longitudinal neuroimaging data

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Early depression detection in young adults is crucial for timely intervention and reducing societal costs.
• Machine learning (ML) offers potential for developing predictive models using complex neuroimaging data.

Purpose of the Study:

• To develop and evaluate ML models for predicting depressive symptoms in young adults.
• To identify neuroimaging features (structural MRI and resting-state fMRI) associated with depression.
• To compare ML-derived features with traditional statistical findings.

Main Methods:

• Longitudinal data including Beck Depression Inventory, sMRI, and rs-fMRI were used.
• Feature selection techniques (LASSO, Boruta, VSURF) identified predictive MRI features.
• Support vector machine and random forest algorithms built prediction models.

Main Results:

• Eight MRI features from specific brain regions (e.g., Orbital Gyrus, Superior Frontal Gyrus) were identified as predictive.
• ML models achieved prediction accuracies from 0.68 to 0.85 and AUCs from 0.57 to 0.81.
• The best model reached 0.85 accuracy and 0.80 AUC, indicating strong predictive potential.

Conclusions:

• Combining sMRI and rs-fMRI with ML shows potential for early depression detection in young adults.
• ML provides unique insights into neural changes associated with depression compared to traditional methods.
• Further research with larger datasets and advanced ML techniques is needed to address overfitting and improve performance.