S2459
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
Purpose/Objective: This study aimed to evaluate the predictive
dosimetric proton-model had AUC 0.955–0.960 versus AUC 0.775–0.776 with dosimetric-only models. BN AUC was 0.75. Vascular comorbidities and baseline visual deficit as key predictors over dosimetric factors. LET and RBE-weighted doses were not associated with RION. Conclusion: Photon-derived NTCP models seem to underperform in proton cohorts and LET parameters were not associated with increased risk of RION. Integrating Bayesian inference with proton-adapted NTCP modeling highlighted the dominant role of vascular and baseline functional factors in determining RION susceptibility, beyond dose-response paradigms. These findings advocate for modality-specific, hybrid predictive frameworks combining clinical and dosimetric parameters to improve individualized risk assessment and guide treatment adaptation. External validation sets are still needed to confirm these findings. Further insights in physics quantities of scanned proton beams are warranted to understand RION at the voxel and functional subunit levels. References: Mayo C, Martel MK, Marks LB, Flickinger J, Nam J, Kirkpatrick J. Radiation Dose–Volume Effects of Optic Nerves and Chiasm. International Journal of Radiation Oncology*Biology*Physics. 2010;76:S28–35.Köthe A, Van Luijk P, Safai S, Kountouri M, Lomax AJ, Weber DC, et al. Combining Clinical and Dosimetric Features in a PBS Proton Therapy Cohort to Develop a NTCP Model for Radiation-Induced Optic Neuropathy. International Journal of Radiation Oncology*Biology*Physics. 2021;110:587–95.Chamseddine I, Shah K, Lee H, Ehret F, Schuemann J, Bertolet A, et al. Decoding Patient Heterogeneity Influencing Radiation-Induced Brain Necrosis. Clin Cancer Res. 2024;30:4424–33. Keywords: modeling, prediction, optic neuropathy MRI-based radiomic and clinical nomograms for personalized risk prediction in locally advanced cervical cancer treated with VMAT and MR-IGABT Wiwatchai Sittiwong 1,2 , Pitchayut Wongsuwan 1,2 , Tissana Prasartseree 1,2 , Wajana Thaweerat 1,2 , Nerisa Thornsri 3 , Pongpop Tuntapakul 1,2 , Pittaya Dankulchai 1,2 1 Division of Radiation Oncology, Department of Radiology, Faculty of Medicine Siriraj hospital, Mahidol University, Bangkok, Thailand. 2 Siriraj Brachytherapy Center (SiBTC), Faculty of Medicine Siriraj hospital, Mahidol University, Bangkok, Thailand. 3 Research Unit, Faculty of Medicine Siriraj hospital, Mahidol University, Bangkok, Thailand Digital Poster Highlight 3139
performance and clinical utility of MRI-based radiomic and clinical nomograms for progression-free survival (PFS) and distant metastasis-free survival (DMFS) in locally advanced cervical cancer (LACC) treated with volumetric modulated arc therapy (VMAT) and MR- guided image-guided adaptive brachytherapy (MR- IGABT). Material/Methods: 140 patients with LACC were included, comprising 100 in the training cohort and 40 in the external testing cohort. All underwent definitive CCRT using VMAT followed by MR-IGABT. MRI-based radiomic and clinical nomograms were developed to predict PFS and DMFS. Radiomic features were extracted from T2WI and DWI of the primary tumor (GTVp) and lymph nodes (GTVn), and significant prognostic variables were identified through univariate and multivariate Cox regression analyses. For external validation, the models were applied to the independent testing cohort, and the Kaplan-Meier method was used to estimate observed 2-year PFS and DMFS rates for comparison with predicted probabilities. Model performance in the external cohort was assessed using discrimination, calibration, and decision curve analysis (DCA). Discrimination was quantified by the C- index, calibration was evaluated with 1,000 bootstrap resamples, and DCA assessed the net clinical benefit of the models across threshold probabilities. Results: Baseline characteristics between the training and testing cohorts were similar. The nomograms for PFS and DMFS incorporated FIGO2018 staging with three radiomic features and T stage at brachytherapy with three radiomic features, respectively. The median predicted 2-year PFS (74.5%) was comparable to the observed 2-year PFS (70.1%), showing good agreement ( Δ =4.4%). Similarly, the predicted and observed 2-year DMFS were nearly identical (77.4% vs. 77.3%). Calibration plots for PFS and DMFS showed close alignment with the 45° reference line, with calibration slopes of 0.90 and 0.98 and intercepts of 0.03 and 0.02, respectively. No statistically significant difference was found between predicted and observed survival (Hosmer–Lemeshow p>0.05). The C-index for the 2- year PFS nomogram was 0.74 (95%CI, 0.68–0.80), and for the DMFS nomogram 0.77 (95%CI, 0.70–0.84), confirming good discrimination and predictive accuracy. DCA demonstrated that both models provided greater net clinical benefit than the “treat-all” and “treat-none” strategies, with the PFS model showing benefit across thresholds of 5–35% and the DMFS model across 5–40%, supporting clinical applicability.
Made with FlippingBook - Share PDF online