S2439
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
Vitaliana De Sanctis 5 , Damiano Caruso 3 , Mattia Falchetto Osti 3 1 UOC Radioterapia, AO San Giovanni Addolorata, Rome, Italy. 2 Department of Surgical and Medical Sciences and Translational Medicine - Radiology Unit, Sant'Andrea University Hospital, Sapienza - University of Rome, Rome, Italy. 3 Department of Medical and Surgical Sciences and Translational Medicine, Faculty of Medicine and Psycology, Sapienza University of Rome, PhD School in Traslational Medicine and Oncology, Rome, Italy. 4 Radiation Oncology Unit, Responsible Research Hospital, Campobasso, Italy. 5 UOC Radioterapia, Sant'Andrea University Hospital, Sapienza - University of Rome, Rome, Italy Purpose/Objective: To develop and validate a machine learning (ML) model based on computed tomography (CT) radiomic features for predicting treatment response in subjects at risk of cancer, under the hypothesis that radiomics can quantify intratumoral heterogeneity associated with disease progression. Material/Methods: In this retrospective observational study, 88 pre - treatment CT datasets were analyzed, including 24 progressive (27.3%) and 64 non - progressive (72.7%) lesions according to post - therapy imaging response. Radiomic features were extracted from segmented lesions to capture intra - lesional heterogeneity. Five supervised ML classifiers—Random Forest, Support Vector Machine, K - Nearest Neighbors, Multi - Layer Perceptron, and Logistic Regression—were trained, validated, and tested for binary classification (“progressive” vs. “non - progressive”). Performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results: The best - performing model achieved an AUC of 0.61 (95% CI 0.51–0.70, p < 0.05), accuracy of 69% (95% CI 64–74, p < 0.005), sensitivity of 36% (95% CI 25–48), specificity of 80% (95% CI 75–86, p < 0.005), PPV of 44% (95% CI 30–57, p < 0.05), and NPV of 78% (95% CI 74– 82, p < 0.005). These results indicate modest but statistically significant discrimination between progressive and non - progressive lesions. Compared with recent multicenter CT - based radiomics models for lung cancer treatment response prediction (AUC 0.75–0.83), the current model shows lower accuracy, likely reflecting limited sample size and single - center design. Conclusion: CT - based radiomic models can non - invasively characterize tumor heterogeneity and partially predict treatment response. Although the present model demonstrated only moderate performance, the
statistically significant classification results support its feasibility as a proof of concept. Expansion to larger, multicenter datasets with standardized radiomic pipelines and inclusion of clinical or genomic variables could enhance predictive accuracy and generalizability. Radiomics thus remains a promising adjunct for individualized therapy planning and risk stratification in oncologic imaging. References: 1.Wu S et al. J Immunother Cancer. 2023; 11(10):e007369.2.Khorrami M et al. Radiology: AI. 2019; 1(2):e180012.3.Trebeschi S et al. Ann Oncol. 2019; 30(6):851 - 859.4.Kothari G et al. Radiother Oncol. 2020; 156:183 - 193. Keywords: SABR, Radiomics, Machine Learning Proffered Paper 2264 From lung to head and neck: exploring effective radiosensitivity as a predictor of treatment response Ali Ameri 1,2 , Simon R. van Kranen 2 , Jan-Jakob Sonke 2 , Olga Hamming-Vrieze 2 , Anna Liza M.P. de Leeuw 2 , Jordi Giralt 3 , Yungan Tao 4 , Sergi Benavente 3 , Thanh-van Nguyen 4 , Frank Hoebers 5 , Ann Hoeben 6 , Catharina H.J. Terhaard 7 , Lip Wai Lee 8 , Signe Friesland 9 , Roel Steenbakkers 10 , Iuliana Toma-Dasu 1 , Marta Lazzeroni 1 1 Department of Physics and Department of Oncology and Pathology, Stockholm University and Karolinska Institutet, Stockholm, Sweden. 2 Department of Radiation Oncology, the Netherlands Cancer Institute, Amsterdam, Netherlands. 3 Department of Radiation Oncology, Hospital General Vall d’Hebron and Vall d’Hebron Institute of Oncology, Barcelona, Spain. 4 Department of Radiation Oncology, Institut Gustave Roussy, Villejuif, France. 5 Department of Radiation Oncology, Maastro Clinic, GROW School for Oncology and Reproduction, Maastricht University Medical Centre+, Maastricht, Netherlands. 6 Division of Medical Oncology, Department of Internal Medicine, GROW School for Oncology and Developmental Biology, Maastricht University Medical Centre+, Maastricht, Netherlands. 7 Department of Radiation Oncology, University Medical Center Utrecht, Utrecht, Netherlands. 8 Department of Clinical Oncology, The Christie NHS Foundation Trust, Manchester, United Kingdom. 9 Department of Oncology and Pathology, and Department of Head, Neck, Lung and Skin Tumors, Karolinska Institutet and Karolinska University Hospital, Stockholm, Sweden. 10 Department of Radiation Oncology, University Medical Center Groningen, Groningen, Netherlands Purpose/Objective: Early assessment of treatment response could facilitate adaptive strategies and enable personalized
Made with FlippingBook - Share PDF online