Integrating CT Foundation Model-Based Imaging Biomarkers and Clinical Characteristics for Progression-Free Survival Prediction In Oligometastatic Renal Cell Carcinoma after Stereotactic Ablative Radiotherapy
Abstract
Purpose
Stereotactic ablative radiotherapy (SAbR) achieves high local control in patients with oligometastatic renal cell carcinoma (omRCC), but a subset of patients experience early progression due to occult micrometastatic disease and may not benefit from an oligometastatic treatment paradigm. Accurate pretreatment risk stratification is therefore critical for identifying patients most likely to benefit from sequential SAbR. This study aimed to develop a prognostic model integrating planning CT-derived imaging biomarkers and clinical characteristics to predict progression-free survival (PFS) in omRCC patients after SAbR.
Methods
We retrospectively analyzed 130 omRCC patients with up to five metastatic sites treated with definitive SAbR between November 2007 and July 2022. Imaging biomarkers were extracted from planning CT using a foundation model, and clinical variables were obtained from the electronic health record. Redundant and irrelevant features were removed by univariate analysis and Spearman rank correlation analysis, followed by feature selection using the minimum Redundancy Maximum Relevance (mRMR) algorithm. Multivariate Cox proportional hazards (CPH) models were constructed using imaging features alone, clinical features alone, and a combined feature set. Model performance was evaluated using five repeats of stratified nested five-fold cross-validation. Risk stratification was assessed using Kaplan-Meier analysis.
Results
The combined model integrating foundation model-based imaging biomarkers and clinical characteristics demonstrated superior prognostic performance compared with single-modality models, achieving a higher mean concordance index (0.658) than models based on imaging features alone (0.596) or clinical features alone (0.631). Using the median predicted PFS expectation as a cutoff, the combined model significantly stratified patients into low- and high-risk groups (log-rank p = 0.001).
Conclusion
Integrating planning CT foundation model-derived imaging biomarkers with clinical characteristics improves prediction of PFS in omRCC patients after SAbR compared with single-modality approaches. This combined framework shows promise for enhancing patient selection, refining risk stratification, and optimizing personalized treatment strategies within the oligometastatic paradigm.