S2431
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
Conclusion: The combined radiomics and dosiomics model demonstrated a superior predictive capability for identifying severe cases of esophagitis compared to clinical variables alone and may enable risk stratification of patients. DVH-based multivariate models may be sufficient for the prediction of radiation-induced esophagitis in clinical settings. References: [1] Bradley J, Forster K. Data from NSCLC-Cetuximab. Published online 2018. doi:10.7937/TCIA.2018.JZE75U7V [2] Seibold, Petra, et al. "REQUITE: a prospective multicentre cohort study of patients undergoing radiotherapy for breast, lung or prostate cancer." Radiotherapy and Oncology 138 (2019): 59-67. Keywords: Radiomics, Esophagitis, Toxicity Prediction Digital Poster 1785 Multi-site trial evaluating FET-PET in Glioblastoma:Central review of FET-PET biologic target volume delineation for adjuvant radiotherapy planning. Eng-Siew Koh 1,2 , Roslyn J. Francis 3,4 , Sze Ting Lee 5,6 , Eddie Lau 6,7 , Elizabeth L Thomas 8 , Angela Whitehead 9 , Olivia Cook 9 , Rachael Dykyj 9 , Alisha Moore 9 , Peter Lin 10 , June Yap 10 , Martin A Ebert 11,12 , Sweet Ping Ng 13 , Mark B Pinkham 14 , Michael Back 15 , Nicholas Bucknell 11 , Isidoro Ruisi 1 , Hui K Gan 16,5 , Andrew M Scott 5,6 1 Radiation Oncology, Liverpool Hospital, Sydney, Australia. 2 South West Sydney Clinical School, University of New South Wales, Sydney, Australia. 3 Nuclear Medicine, Sir Charles Gairdner Hospital, Perth, Australia. 4 Medical School, The University of Western Australia, Perth, Australia. 5 Tumour Targeting Program, Olivia Newton-John Cancer Research Institute, Melbourne, Australia. 6 Department of Molecular Imaging and Therapy, Austin Health, Melbourne, Australia. 7 Department of Radiology, Austin Health, Melbourne, Australia. 8 Department of Nuclear Medicine, Sir Charles Gairdner Hospital, Perth, Australia. 9 TROG Cancer Research, TROG, Newcastle, Australia. 10 Department of Nuclear Medicine, Liverpool Hospital, Sydney, Australia. 11 Department of Radiation Oncology, Sir Charles Gairdner Hospital, Perth, Australia. 12 School of Physics, Mathematics and Computing, The University of Western Australia, Perth, Australia. 13 Department of Radiation Oncology, Austin Health, Melbourne, Australia. 14 Department of Radiation Oncology,, Princess Alexandra Hospital and ICON Cancer Centre, Brisbane, Australia. 15 Department of Radiation Oncology, Royal North Shore Hospital,, Sydney, Australia. 16 Department of Medical Oncology,, Austin Health, Melbourne, Australia
Tomek. In a subsequent cross-validation of the training data, the 33rd and 66th percentiles of predicted probabilities were used as thresholds to categorize patients into risk groups, which were then evaluated on the external test set. Differences in incidence rates between these groups were assessed using Fisher's exact test.
Results: All dosiomics models outperformed the clinical model and random chance (ROC-AUC > 0.5). The best model within the nCV was based on a combination of radiomics and physical dosiomics, achieving a mean ROC-AUC of 0.65 in training and 0.76 (95% CI: 0.49-1) during testing. The DosiomicsPhysical and DVH-based models showed similar performance in the nCV (ROC- AUC: 0.63), achieving ROC-AUCs of 0.81 (95% CI: 0.62- 1) and 0.83 (95% CI: 0.66-1) on the test set, respectively. Stratification of patients revealed an incidence rate of less than 1% in the low-risk group, with a trend towards significance (p = 0.052) observed through Fisher's exact test.
Made with FlippingBook - Share PDF online