ESTRO 2025 - Abstract Book

S3359

Physics - Machine learning models and clinical applications

ESTRO 2025

134

Mini-Oral Identifying the optimal timepoint for adaptive re-planning in prostate cancer radiotherapy to minimise rectal toxicity using normal tissue radiomics Zhuolin Yang 1,2,3 , David J Noble 3,4 , Sarah Elliot 1,3,5 , Leila Shelley 1 , Thomas Berger 1 , Raj Jena 6 , Duncan B McLaren 3,4 , Neil G Burnet 7 , William H Nailon 1,2,3 1 Department of Oncology Physics, Edinburgh Cancer Centre, Western General Hospital, Edinburgh, United Kingdom. 2 Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh, Edinburgh, United Kingdom. 3 Edinburgh Cancer Research, CRUK Scotland Centre, Institute of Genetics and Cancer, University of Edinburgh, Edinburgh, United Kingdom. 4 Department of Clinical Oncology, Edinburgh Cancer Centre, Western General Hospital, Edinburgh, United Kingdom. 5 Centre for Medical Informatics, Usher Institute, University of Edinburgh, Edinburgh, United Kingdom. 6 Department of Oncology, The University of Cambridge, Cambridge, United Kingdom. 7 Proton Beam Therapy Centre, The Christie NHS Foundation Trust, Manchester, United Kingdom Purpose/Objective: Adaptive radiotherapy allows for dynamic adjustments during treatment, improving tumour targeting and reducing radiation exposure to healthy tissues. This study aims to determine the optimal timing for adaptive re-planning by utilising radiomic features from pre-treatment CT and daily MVCT image guidance scans, with the specific goal of minimising rectal bleeding. Material/Methods: This analysis utilised data from the VoxTox study (UK-CRN-ID-13716), including 187 prostate cancer patients who underwent TomoTherapy with daily image guidance [1]. Patients were categorised into two groups based on their prescribed doses: 110 received 74 Gy in 37 fractions (7½ weeks), and 77 received 60 Gy in 20 fractions (4 weeks). Grade ≥ 1 rectal bleeding was evaluated two years post -radiotherapy based on CTCAE v4.03. IBSI-compliant radiomic features were extracted from 8x8 pixel 2 subimages on the rectal wall on planning CTs and daily MVCTs, and subsequently averaged across subimages and slices (Fig.1) [2]. For MVCTs, radiomic features were further averaged weekly to generate week-level features. The dataset was divided into training (75%, N = 139) and test (25%, N = 48) sets. Radiomic features from different timepoints were analysed separately and cumulatively to investigate their individual and combined power for predicting rectal bleeding. Mann-Whitney U tests and Spearman’s rank correlation were used to eliminate redundant predictors, followed by multivariate logistic regression with elastic net penalty, fine-tuned by 5-fold cross-validation. By assessing the predictive performance of radiomic features at multiple timepoints, the most informative timepoint for adapting treatment plans was identified.