ESTRO 2025 - Abstract Book

S3863

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Keywords: NSCLC, causal inference, counterfactual prediction

References: [1] W. A. C. van Amsterdam et al. , ‘Individual treatment effect estimation in the presence of unobserved confounding using proxies: a cohort study in stage III non-small cell lung cancer’, Sci Rep , vol. 12, no. 1, p. 5848, Dec. 2022, doi: 10.1038/s41598-022-09775-9. [2] A. Aupérin et al. , ‘Meta-Analysis of Concomitant Versus Sequential Radiochemotherapy in Locally Advanced Non–Small-Cell Lung Cancer’, JCO , vol. 28, no. 13, pp. 2181–2190, May 2010, doi: 10.1200/JCO.2009.26.2543.

4438

Digital Poster Artificial intelligence and radiomics-based models to predict clinical response to low-dose radiotherapy in arthrodegenerative pathology. Paula Maixé Brull 1,2 , Marta Canela Capdevila 2 , Raquel Garcia Pablo 2 , Alberto Martínez Caballero 3 , Judith Osuna Ramírez 1,2 , Yolanda López Sánchez 1 , Andrea Jiménez Franco 2 , Juan Manuel Jiménez Aguilar 2 , Rocío Benavides Villarreal 1,2 , Jordi Camps Andreu 2 , Meritxell Arenas Prat 2 , Víctor Hernández Masgrau 4 1 Department of Radiation Oncology, Hospital Universitari de Sant Joan, Reus, Spain. 2 Unitat de Recerca Biomèdica, Institut d'Investigació Sanitària Pere Virgili, Reus, Spain. 3 Medical Image Analysis and Biometry Laboratory, Universidad Rey Juan Carlos, Madrid, Spain. 4 Department of Medical Physics, Hospital Universitari de Sant Joan, Reus, Spain Purpose/Objective: This study aims to develop and validate artificial intelligence and radiomics-based predictive models of clinical response of patients with arthrodegenerative hand disease treated with low-dose radiotherapy (LDRT). Material/Methods: 92 patients with hands osteoarthritis and treated with LDRT were analysed. Demographic and treatment data were collected from medical records and 1689 radiomic features were extracted from simulation computed tomography (CT) images. To select the most relevant features, Mutual Information, Correlation, Random Forest (RF) Supervised Selection, ANOVA and LassoCV methods were combined. Finally, several machine learning models were trained and validated with cross validation, 80% of the data were used for training and 20% for testing. These models were Random Forest, Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel and Convolutional Neural Networks (CNN). Results: The mean age of the patients was 64.7 years, 90.2% of whom were women. 69.6% of clinical responses were positive, while the remaining 30.4% of patients did not respond to LDRT. Predictive models based on 12 radiomic features obtained a mean area under the curve (AUC) score of 0.73 for RF, 0.76 for SVM and 0.73 for CNN. Confusion matrices for the validation test for the SVM with RBF kernel produced 10 True Positives and 9 True Negatives, with 3 False Positives and 4 False Negatives, showing few errors. The RF correctly classified 10 positives and 9 negatives, but gave 4 False Positives and 3 False Negatives. The CNN scored 8 True Positives and 10 True Negatives and had more False Negatives (5). SVM and CNN demonstrated superior specificity (0.769 for both) but lower sensitivity compared to RF (SVM: 0.692, CNN: 0.615). RF achieved a sensitivity of 0.769 and a specificity of 0.692.