Therapy Response Prediction in Patients with Metastatic Soft Tissue Sarcomas Using CT based Delta Radiomics

This retrospective study investigates whether computed tomography (CT)-based delta-radiomics can improve systemic treatment response prediction in patients with metastatic STS. Data from 71 patients with initially unresectable, high-grade, metastasized STS treated at Erasmus Medical Center between 2014 and 2020 were included. Radiomics features were extracted from up to five metastases, and delta-radiomics were computed as the relative difference in features between pre-treatment and follow-up scans. Therapy response was modeled using survival analysis, utilizing the time interval from metastasis diagnosis to death or latest follow-up. To predict response, we employed automated machine learning differentiating three input configurations: 1) The imaging model, based on 107 quantitative features; 2) a diameter-only model; and 3) a volume-only model. Models were evaluated using a repeated nested 5-fold cross-validation. The imaging model achieved a mean c-index of 0.68 (95% CI: 0.60–0.76) and a one-year cumulative dynamic area under the curve (cAUC) of 0.75 (95% CI: 0.55–0.95). Diameter- and volume-only models performed worse, with c-indices of 0.61 (95% CI: 0.51–0.70) and 0.65 (95% CI: 0.53–0.76), and cAUCs of 0.60 (95% CI: 0.35–0.85) and 0.63 (95% CI: 0.38–0.88), respectively. These findings suggest that CT-based delta-radiomics is valuable for predicting therapy response in metastatic STS, warranting potential optimization and validation in larger, multi-center studies before clinical adoption.