ObjectiveThe aim of this study was to develop a pretreatment magnetic resonance imaging (MRI)-based radiomics model for disease-free survival (DFS) prediction in patients with uveal melanoma (UM).MethodsWe randomly assigned 85 patients with UM into 2 cohorts: training (n = 60) and validation (n = 25). The radiomics model was built from significant features that were selected from the training cohort by applying a least absolute shrinkage and selection operator to pretreatment MRI scans. Least absolute shrinkage and selection operator regression and the Cox proportional hazard model were used to construct a radiomics score (rad-score). Patients were divided into a low- or a high-risk group based on the median of the rad-score. The Kaplan-Meier analysis was used to evaluate the association between the rad-score and DFS. A nomogram incorporating the rad-score and MRI features was plotted to individually estimate DFS. The model's discrimination power was assessed using the concordance index.ResultsThe radiomics model with 15 optimal radiomics features based on MRI performed well in stratifying patients into the high- or a low-risk group of DFS in both the training and validation cohorts (log-rank test, P = 0.009 and P = 0.02, respectively). Age, basal diameter, and height were selected as significant clinical and MRI features. The nomogram showed good predictive performance with concordance indices of 0.741 (95% confidence interval, 0.637-0.845) and 0.912 (95% confidence interval, 0.847-0.977) in the training and validation cohorts, respectively. Calibration curves demonstrated good agreement.ConclusionThe developed clinical-radiomics model may be a powerful predictor of the DFS of patients with UM, thereby providing evidence for preoperative risk stratification.