ESTRO 2025 - Abstract Book

S3775

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Figure 1: Comparison between the selected plans. Meaning of abbreviations: S = Siemens, VR = Varian, E = Elekta, 3D= 3D CRT, IM= IMRT d= forward planning, i = inverse planning, 6= 6 MV

Conclusion: Shuryak’s model, which accounts for dose fractionation and interfractionary repair, successfully identified significant differences in ERR between various prostate radiotherapy plans with equivalent biological effects. These findings highlight the importance of incorporating secondary cancer risk estimation into the treatment plan optimization process. By integrating such predictive models into clinical practice, radiotherapy treatments could be further refined to enhance long-term patient safety and reduce the risk of secondary malignancies.

Keywords: Radiobiology model, peripheral dose, second cancer

References: [1] Shuryak, I. et al. doi: 10.1007/s00411-009-0230-3 [2] Shuryak, I. et al. doi: 10.1007/s00411-009-0231-2 [3] Dasu, A. et al. doi: 10.1016/j.ejmp.2017.02.015 [4] Sánchez-Nieto B et al. doi: 10.3389/fonc.2022.872752 [5] Muñoz-Hernández IS et al. doi: 10.1016/j.ejmp.2023.103183 [6] Sánchez-Nieto B et al. doi: 10.1016/j.ejmp.2019.09.076

2283

Digital Poster Machine Learning-Based Prediction of Dermatitis in Hypofractionated Breast Radiotherapy Patients: Combined Clinical, Radiomic, and Dosiomic Analysis Yen-Ting Liu 1,2,3 , Ting-An Chang 4 , Yung-Chieh Li 4 , Kuan-An Chu 2 , Shih-Yin Chen 5 , Shih-Ting Huang 2 1 Biomedical Engineering, National Taiwan University, Taipei, Taiwan. 2 Oncology, National Taiwan University Hospital Yunlin Branch, Yunlin, Taiwan. 3 Oncology, National Taiwan University Hospital, Taipei, Taiwan. 4 Electrical Engineering, National Yunlin University of Science and Technology, Yunlin, Taiwan. 5 Cancer Registry, National Taiwan University Hospital Yunlin Branch, Yunlin, Taiwan Purpose/Objective: Breast cancer has emerged as the leading cancer worldwide in women, accounting for 11.7% of newly diagnosed cases. Radiation-induced dermatitis following hypofractionated breast radiotherapy remains a critical concern, potentially leading to treatment interruptions, reduced adherence, and compromised outcomes. This study aims to develop and validate machine learning models for predicting radiation-induced dermatitis in breast cancer patients treated with hypofractionated radiotherapy. Material/Methods: This IRB-approved retrospective study, conducted from March 2017 to December 2022, included patients with breast cancer receiving breast conservative treatment followed by hypofractionated radiotherapy (4005–5605 cGy in 15–23 fractions). Patient data were collected from the institutional cancer registry. Using 3D Slicer and Pyradiomics, features were extracted from whole breast and tumor bed PTV (planning target volume) regions. The analysis incorporated clinical parameters, radiomics and dosiomics features from simulation and planning CT images. Dermatitis was graded according to CTCAE 5.0. Multiple prediction models combining structured clinical data, imaging and dosiomics features were developed. Results: From an initial cohort of 193 patients, 124 met inclusion criteria after data cleaning. The final dataset comprised 20 clinical, 112 radiomics, and 224 dosiomics features for each patient. We divided the patient data into training and