S1854
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
Standard IDOL [2], providing precision without uncertainty estimation; SingleP [4], training solely on individual patient data without pre-training; and general model (baseline).All methods were evaluated on adapted treatment plans of 21 test patients for dose prediction accuracy. For IDOL_CTr, dose error estimation was investigated through correlation analysis between CT reconstruction error and DP error for each model. Results: Figure1 shows dose metric errors relative to manual plans for different strategies. Median errors for IDOL_CTr and IDOL were consistently lower than the general model, except for optical nerves and eyes where absolute median errors were already low ( ≤ 1.1Gy). IDOL_CTr's largest median difference from IDOL was 1.35Gy. SingleP was consistently outperformed, with the largest median difference reaching 17.1Gy for mandible versus the general model.
Conclusion: IDOL_CTr successfully combined high-precision PS dose prediction with integrated uncertainty quantification essential for safe ART implementation. While maintaining accuracy comparable to standard IDOL, the method provided uncertainty measures through reconstruction error, advancing clinical viability for real-time adaptive decision-making. References: [1] D. Nguyen et al., "A comparison of Monte Carlo dropout and bootstrap aggregation on the performance and uncertainty estimation in radiation therapy dose prediction with deep learning neural networks," Phys. Med. Biol., vol. 66, no. 5, p. 054002, 2021.[2] A. Maniscalco et al. "Intentional deep overfit learning for patient - specific dose predictions in adaptive radiotherapy," Med. Phys., vol. 50, no. 9, pp. 5354-5363, 2023.[3] M. Huet - Dastarac et al., "Can input reconstruction be used to directly estimate uncertainty of a dose prediction U - Net model?," Med. Phys., 2024.[4] A. Maniscalco et al. , "Single patient learning for adaptive radiotherapy dose prediction," Med. Phys., vol. 50, no. 12, pp. 7324-7337, 2023. Keywords: adaptive, patient-specific, uncertainty estimation Redefining pelvic dose acceptance thresholds: a ROC-based dosimetric analysis of gastrointestinal adverse events in high-risk endometrial cancer. Raouia Ben Amor 1,2 , Ines Mlayeh 1,2 , Roua Toumi 1 , Zeineb Naimi 1,2 , Rihab Haddad 1 , Ghada Bouguerra 1 , Lotfi Kochbati 1,2 1 Radiation Oncology, Abderrahmen Mami Hospital, Ariana, Tunisia. 2 Faculty of Medicine of Tunis, Tunis El Manar University, Tunis, Tunisia Purpose/Objective: The combined use of adjuvant chemotherapy (AdCT)and whole-pelvis radiotherapy (WPRT) has Digital Poster 1715
Figure2 illustrates the correlation between CT reconstruction error and DP error for two IDOL_CTr models. The different models slopes demonstrate the benefit of patient-specific calibration to estimate DP error. Following a linear fit for each model, estimated median dose errors for adapted plans were 0.8 and 1.3Gy for models 1 and 2, respectively, while actual values were 1.2 and 1.5Gy.
Made with FlippingBook - Share PDF online