S1941
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
dose (5.4% ± 6.4%, p<0.05). Planning time dropped from 2–3 days to ~3–5 hours, and 77% of plans were rated as equivalent or superior by physicians.For VMAT-CSI, auto-plans demonstrated significant OAR sparing, particularly to the parotids, submandibular glands, and thyroid, and decreased body V50%, with 88.3% of plans rated as equivalent or better. Planning time was reduced from ~5–6 hours to 1–2 hours.For VMAT-TMLI, auto-plans preserved target coverage and heterogeneity while improving sparing of the kidneys, larynx, thyroid, and oral cavity. The consistency of DVH metrics was higher, and planning time was reduced from 2–3 days to ~6 hours. Conclusion: This unified scripting solution enables efficient, high- quality planning for VMAT-TBI, CSI, and TMLI. The streamlined workflow improves consistency, reduces planner workload, and broadens patient access to complex radiotherapy. The VMAT-TBI and CSI tools are now routinely used at our institution, with over 175 TBI and 25 CSI patients treated. All scripts are open-
efficient dose prediction in breast radiotherapy. Although the model exhibited lower target-dose accuracy than sequential models, this may reflect the influence of anatomical coupling inherent to multi-task training. Future work will focus on improving segmentation accuracy for target structures to enhance the overall dose prediction performance. References: [1] Li, H., Peng, X., Zeng, J., Xiao, J., Nie, D., Wang, Y., 2022. Explainable attention guided adversarial deep network for 3D radiotherapy dose distribution prediction. Knowl. Based Syst. 241. Keywords: Multi-task learning, Dose prediction, Segmentation Poster Discussion 3576 An Automated VMAT Planning Framework for TBI, CSI, and TMLI: Enhancing Efficiency and Plan Quality Eric Simiele 1,2 , Ignacio Romero 3,2 , Caressa Hui 4,2 , Michael Binkley 2 , Iris Gibbs 2 , Richard Hoppe 2 , Susan M Hiniker 2 , Nataliya Kovalchuk 2 1 Radiation Oncology, 2The University of Alabama at Birmingham, Birmingham, USA. 2 Radiation Oncology, Stanford University, Stanford, USA. 3 Radiation Oncology, Fresno Cancer Center, Fresno, USA. 4 Radiation Oncology, University of California, Irvine, USA Purpose/Objective: We present a fully automated, open-source VMAT planning framework that supports three complex radiotherapy techniques: Total Body Irradiation (VMAT-TBI), Craniospinal Irradiation (VMAT-CSI), and Total Marrow and Lymphoid Irradiation (VMAT-TMLI). This unified framework aims to streamline planning, improve consistency, and maintain or enhance plan quality across all modalities. The script is publicly available via GitHub to support global implementation. Material/Methods: The framework was built using the Varian Eclipse Scripting Application Programming Interface to automate all three planning modalities. A validation cohort of 35 patients (10 TBI, 20 CSI, 5 TMLI) was used to compare auto-plans versus clinical manual plans. Plans were evaluated using DVH-based target and OAR metrics, as well as blinded physician review. Efficiency gains were assessed by tracking planning time. Following validation, the tool was adopted clinically, and the VMAT-TBI component was incorporated into the multi-institutional COG ASCT2031 trial. Results: For VMAT-TBI, automated plans achieved comparable or improved PTV coverage, lower Dmax, and reduced kidney dose, with a significant reduction in mean lung
source and actively maintained, allowing for integration into diverse clinical environments worldwide. Keywords: VMAT-TBI, VMAT-CSI, VMAT-TMLI
Digital Poster 3587 Electron density structure overrides for brain treatments on the MR-Linac: How many structures do we really need? Sandra Fisher 1,2 , Ben Starvaggi 1 , Tegan Courtot 1 , Leah McDermott 1 , Morikatsu Wada 1 , Ricky O'brien 3 , Sweet Ping NG 1,2 1 Department of Radiation Oncology, Austin Health, Melbourne, Australia. 2 Department of Surgery, University of Melbourne, Melbourne, Australia. 3 Medical Radiations, School of Health and Biomedical Science, RMIT University, Melbourne, Australia Purpose/Objective: MR datasets do not contain tissue density information in a format that treatment planning system dose calculation algorithms can use for calculations. On the Elekta Unity MR- Linac this is overcome by assigning bulk density overrides to a user defined number of structures to represent the range of tissue densities in the dose calculation grid. This work investigated the impact on the dose calculation that varying the number of override structures and their assigned densities had for a cohort of brain patients planned for treatment on the MR-Linac. Material/Methods: A retrospective planning study was conducted on 10 glioblastoma and 10 brain metastasis patients treated on the MR-Linac at the Austin Health, Melbourne,
Made with FlippingBook - Share PDF online