Personalized prediction of response to selective serotonin reuptake inhibitors (SSRIs) treatment for major depression disorder using deep learning networks from connectivity maps of EEG

Accurate prediction of antidepressant treatment response remains an important challenge in depression research. In this preliminary exploratory study, we propose an electroencephalogram (EEG)-based deep learning framework for predicting response to selective serotonin reuptake inhibitors (SSRIs) in patients with major depressive disorder (MDD). Multichannel resting-state EEG signals were analyzed using functional and effective connectivity measures, including Phase Lag Index (PLI), Phase Locking Value (PLV), Partial Directed Coherence (PDC), and Directed Transfer Function (DTF). These measures were fused across five frequency bands to construct fused frequency connectivity images (FFCIs), combining complementary aspects of functional and effective brain connectivity. Multiple deep learning architectures, including ConvNeXt, Swin Transformer, EfficientNet-B0/B3, MobileNetV3, ResNet-18, and MambaVision, were evaluated using subject-independent cross-validation on the Mumtaz EEG database. MobileNetV3 achieved the highest observed patient-level accuracy of 78.78%, followed by MambaVision at 75.75%. Controlled fusion comparisons, representation-level ablation analyses, and a classical RBF-SVM baseline using numerical connectivity features were additionally conducted to assess the contribution of the proposed representation. Paired model comparisons using exact McNemar tests showed that the observed performance differences did not remain statistically significant after correction for multiple comparisons. Integrated Gradients provided exploratory insights into the connectivity patterns contributing to model predictions. Overall, the results demonstrate the feasibility of combining fused connectivity representations with modern deep learning models for EEG-based prediction of SSRI treatment response; however, the findings require validation in larger, independent cohorts before clinical application.