S2427
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
significantly associated with the risk of hs-cTnT elevation for the pre-practice change cohort (odds ratio (OR):1.045 (95%CI:1.005-1.088), p = 0.029) but not for the post-practice change cohort (OR: 1.026 (95%CI: 0.950-1.106), p = 0.499). Bayesian inference revealed that the probability of OR > 1 decreased post- implementation, with a 63.8% probability that the MHD effect was reduced.Notably, the univariable and multivariable pLR models identified a dose threshold of 12.7 Gy. Below this threshold, hs-cTnT elevation risk showed minimal MHD dependence; above it, risk increased substantially with MHD. The pLR models outperformed the cLR models in overall performance (AIC: 278-287 vs 282-290) and discriminative ability (AUC 0.64-0.71 vs 0.58-0.68).LRT p-values of 0.022 confirmed that pLR models significantly improved model fit compared with cLR models.
Conclusion: We observed a non-linear, piecewise dose-response relationship of MHD on the risk of cardiac troponin elevation with complication probability independent of MHD up to 12.7Gy, but increases thereafter. External validation in prospective cohorts is warranted. Keywords: piecewise, dose threshold, cardiac risk
Proffered Paper 1216
Limited reproducibility of published radiomics models in rectal cancer: Results from an external validation Lukas Dünger 1 , Emily Mäusel 1 , Christian Fischer 1,2 , Annika Lattermann 1,2 , Mechthild Krause 1,2 , Esther Troost 1,3 , Alex Zwanenburg 1,4 , Steffen Löck 1,5 1 OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Dresden, Germany. 2 Department of Radiotherapy and Radiation Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Dresden, Germany. 3 Helmholtz-Zentrum Dresden- Rossendorf, Institute of Radiooncology – OncoRay, Dresden, Germany. 4 National Center for Tumor Diseases, NCT/UCC, Dresden, Germany. 5 German Cancer Consortium, Partner Site Dresden, Heidelberg, Germany Purpose/Objective: Radiomics has been proposed as a key technology to advance precision oncology by transforming medical images into quantitative biomarkers for diagnosis and prognosis. Despite a substantial rise in published studies, radiomics has seen minimal clinical adoption. A major reason is the lack of reproducibility and external validation, which undermines confidence in its clinical reliability. Addressing this gap is critical for moving the field beyond proof-of-concept research toward real-world utility. This study performed a
Made with FlippingBook - Share PDF online