Advancing Early Detection of Major Depressive Disorder Using Multisite Functional Magnetic Resonance Imaging Data:...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Major depressive disorder (MDD) is a widespread mental health issue with significant public health consequences.
• Current MDD diagnosis relies on subjective assessments, often causing delays or errors.
• Neuroimaging and machine learning (ML) offer potential for objective and accurate early MDD detection.

Purpose of the Study:

• To develop and validate ML models for early MDD detection using multisite functional magnetic resonance imaging (fMRI) data.
• To compare the performance of different ML models.
• To evaluate the clinical applicability of these models for early MDD detection.

Main Methods:

• Utilized fMRI data from 1200 participants (600 with early-stage MDD, 600 controls) across three public datasets.
• Trained and evaluated four ML models (SVM, Random Forest, GBM, DNN) using 5-fold cross-validation.
• Assessed model performance using accuracy, sensitivity, specificity, F1-score, and AUC; employed SHAP and activation maximization for interpretation.

Main Results:

• The deep neural network (DNN) model achieved 89% accuracy and 0.95 AUC, significantly outperforming traditional methods.
• Key predictors included altered functional connectivity in prefrontal cortex, anterior cingulate cortex, and limbic regions.
• The DNN model showed 78% sensitivity in predicting MDD development within a 2-year follow-up and demonstrated cross-dataset generalizability.

Conclusions:

• AI-driven approaches using fMRI show significant potential for early MDD detection and intervention.
• These AI tools should augment, not replace, clinical judgment.
• Ethical considerations, including patient privacy and model bias, require careful attention.