Advancing Early Detection of Major Depressive Disorder: A Comparative Analysis of AI Models Using Multi-Site Functional MRI Data

Background Major Depressive Disorder (MDD) is a prevalent mental health condition with significant public health implications. Early detection is crucial for effective intervention, yet current diagnostic methods often fail to identify MDD in its early stages. Objective This study aimed to develop and validate machine learning models for the early detection of MDD using functional Magnetic Resonance Imaging (fMRI) data. Methods We utilized fMRI data from 1,200 participants (600 with early-stage MDD and 600 healthy controls) across three public datasets. Four machine learning models (Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Machine (GBM), and Deep Neural Network (DNN)) were developed and compared. Models were evaluated using accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC-ROC), and F1 score. Results The DNN model demonstrated superior performance, achieving 89% accuracy (95% CI: 0.86-0.92) and an AUC-ROC of 0.95 (95% CI: 0.93-0.97) in detecting early-stage MDD. Key predictive features included altered functional connectivity between the dorsolateral prefrontal cortex, anterior cingulate cortex, and limbic regions. The model showed good generalizability across different datasets and identified 78% (95% CI: 71%-85%) of individuals who developed MDD within a 2-year follow-up period. Conclusions Our AI-driven approach demonstrates promising potential for early MDD detection, outperforming traditional diagnostic methods. This study highlights the utility of machine learning in analyzing complex neuroimaging data for psychiatric applications. Future research should focus on prospective clinical trials and the integration of multimodal data to enhance the clinical applicability of this approach further.