Progression-Free Survival Prediction Using Computed Tomography Foundation Models In Stereotactic Ablative Radiotherapy–Treated Oligometastatic Renal Cell Carcinoma
Abstract
Purpose
Stereotactic ablative radiotherapy (SAbR) has demonstrated efficacy in controlling oligometastatic renal cell carcinoma (omRCC) with safely delayed systemic therapy. However, a subset of patients has limited benefit from SAbR, which may require upfront systemic therapy to control the risk of metastatic progression. Early stratification of poor- versus good-responding patients could therefore support more individualized treatment decision-making for this patient population.
Methods
We developed a model to predict progression-free survival (PFS) in a cohort of 130 patients with omRCC treated with SAbR between November 2007 and July 2022. Patient demographic, clinical, and treatment-related variables were collected, along with PFS outcomes. Planning CT images were available for all patients. A pretrained foundation model for 3D CT image analysis (Merlin) was used as the image feature extractor. Imaging and clinical features were integrated using a multitask logistic regression model to predict PFS risk scores. Two training strategies were evaluated: (1) using the foundation model solely as a fixed feature extractor without fine-tuning, and (2) fine-tuning the foundation model on this patient cohort. The dataset was randomly divided into training (60%), validation (20%), and testing (20%) sets.
Results
When the foundation CT model was used without fine-tuning, the concordance index (C-index) on the testing set was 0.53. Fine-tuning the foundation model on the omRCC cohort improved predictive performance, increasing the C-index to 0.69.
Conclusion
Task- and dataset-specific fine-tuning appears to be preferable when applying foundation models for imaging feature extraction. The proposed fine-tuned survival prediction model may enable earlier identification of patients unlikely to benefit from SAbR alone, allowing timely integration of systemic therapies alongside local treatment to potentially improve outcomes in patients with omRCC.