ESTRO 2025 - Abstract Book

S3804

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

2935

Digital Poster Deep Radiomic Analyses of Genomics and Oncologic Scans (DRAGONs) Pritha Roy 1 , Richard Alexander Adams 1 , Emiliano Spezi 2 , Concetta Piazzese 2 1 Clinical Oncology, Velindre NHS Trust, Cardiff, United Kingdom. 2 Medical Engineering, Cardiff University, Cardiff, United Kingdom Purpose/Objective: Rectal cancer therapy should be tailored to the individual. To best support patients in their decision-making, it is important to identify those who might benefit from de-escalation strategies including non-operative management. TNM staging with MRI pelvis remains the most effective indicator of prognosis but limited data from the image is used to support clinical decisions. Here we explore the use of image based signatures (radiomics) associated with prognosis (phase 1), and predicting response to total neo-adjuvant therapy (phase 2). Material/Methods: All patients receiving neo-adjuvant RT in a tertiary cancer centre between January 2010 - March 2019 were included in phase 1. Gross tumor volume was delineated in baseline MRIs for each patient, transferred to SPAARC (Spaarc pipeline for automated analysis and radiomics computing) software where shape and texture extraction was performed and 299 relevant radiomic features were identified. Four most reflective radiomics features were identified and combined with 9 clinical variables (Table) for evaluation. The model created in SPSS with these parameters converged after 10 steps and only the radiomic variable Auto Correlation GLCM (AC_GLCM) was found to be significant and explored further. Patients were divided into low (LR), intermediate (IR) and high (HR) risk groups based on the AUC for AC_GLCM. Overall survival (OS) was determined by Kaplan-Meier plot for each risk group. Results: A total of 235 patients were analysed, of which 162 (68.9%) patients met the eligibility criteria. Demographic features of the cohort are in Table . The median follow-up was 40 months. The median OS was not reached however, OS at 2 years and 3 years were 90.2% and 84.7% respectively. On log-rank test there was a non-significant difference (p 0.095) in OS between the groups however, on merging the LR and IR groups there was statistically significant difference in OS (p 0.032). (Figure)