On the value of radiomics in addition to clinical measures in emotional conflict fMRI for predicting sertraline response in major depressive disorder.

Most patients with major depressive disorder (MDD) do not respond to first-line treatment, and robust predictors of individual outcomes are lacking. Functional MRI (fMRI) has shown promise for identifying treatment-relevant brain markers, but standard analyses often rely on mean activation within regions of interest (ROIs). Emotional-conflict fMRI elicits spatially heterogeneous, potentially clinically relevant activation, which radiomics can capture by quantifying the texture and distribution of voxel-wise activity. In this preregistered study, we tested whether radiomic features improved the prediction of treatment response compared to mean fMRI ROI activation at baseline (week 0) and early-treatment (week 1) of 254 MDD outpatients participating in the EMBARC RCT (123 sertraline, 131 placebo). Across models, clinical variables provided the strongest predictive signal, and imaging-only models showed limited predictive utility. Placebo but not sertraline response could be predicted at week 1 early-treatment with three combined models incorporating different levels of imaging information, although not at week 0 baseline. In our regression analysis, combined models including mean ROI features yielded modest performance for sertraline (baseline: r = 0.22; early-treatment: r = 0.29; P < 0.05), whereas radiomic models underperformed (baseline: P > 0.05), whereas whole-brain models modestly improved prediction (baseline: r = 0.29; P = 0.0012). Early clinical measures remained the strongest predictors with higher importance than radiomic features. These findings suggest limited added value of emotional conflict fMRI or the addition of radiomics features over clinical data for predicting treatment outcomes in MDD.