S1856
Physics - Dose prediction/calculation, optimisation and applications for photon and electron planning
ESTRO 2026
Proffered Paper 1804
using OrthoPlan were run using EGSnrc and the BEAMnrc and DOSXYZnrc user libraries. 3D Slicer (v5.8.1) was used for override contouring, Monte Carlo dose calibrations, and figure generation. Results: The handbook currently features 15 retrospective dose calculations, covering the nose, cheek, canthus, ears, forehead and chest, shown alongside diagnosis, prescription and treatment information, photographs of treatment areas, dose quality metrics (such as surface dose homogeneity), and dose distributions optimised for other modalities where available. Notes describing uncertainties in Monte Carlo modelling and calculation, including dose grid resolution, voxelization and volume averaging, tissue composition, treatment set-up, and use of dose-to-medium, are also included.Figure 1 shows the superficial radiotherapy dose calculated for a nasal bridge basal cell carcinoma patient, along with the corresponding volumetric modulated arc therapy (VMAT) calculation, illustrating a clear difference in ocular dose from the two treatments. This treatment was planned for delivery with a 100 kV, 2.90 mm Al half-value-layer beam with a 5 cm applicator.
Towards LLM-based universal automated treatment planning solution for photon radiotherapy: Application in locally advanced rectal cancer VMAT planning Jiacheng Liu 1,2 , Kaining Yao 2 , Meijiao Wang 2 , Tong Zhang 3 , Xiaoyu Yang 3 , Yichen Pu 2 , Haizhen Yue 2 , Yi Du 2,3 , Hao Wu 2,3 , Ruoxi Wang 2 , Lisheng Geng 1 1 School of Physics, Beihang University, Beijing, China. 2 Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Department of Radiation Oncology, Peking University Cancer Hospital & Institute, Beijing, China. 3 Institute of Medical Technology, Peking University Health Science Center, Beijing, China Purpose/Objective: The increasing workload of radiotherapy planning and personnel shortages underscore the urgent need for efficient and generalizable automated treatment planning (ATP) methods [1]. Current data-driven ATP methods are often limited by the scarcity of high- quality planning data and exhibit inadequate generalization when faced with diverse clinical prescriptions [2-3]. To address these challenges, this study introduces a universal text-driven ATP framework, UText-Planner, based on large language models (LLM), for locally advanced rectal cancer (LARC). The framework aims to distill expert planning knowledge to support flexible and prescription- adaptive treatment planning. Material/Methods: ΠΆ he UText-Planner differs from conventional data- driven approaches by integrating an LLM that responds to textual instructions, as is illustrated in Figure 1. This integration enables three core functions: (1) distilling expert planning knowledge via few-shot learning, (2) constructing a universal adapter to accommodate varying clinical prescriptions, and (3) guiding the inverse planning process through text- based knowledge via API calls to a clinical TPS - specifically, Eclipse, in this study.The UText-Planner was developed through multiple iterations of planning data from three LARC patients and initial planning objectives from five representative cases. Its feasibility and generalizability were evaluated on a cohort of 75 patients, including an internal validation set (n = 30) with prescriptions consistent with the training cases, and an external validation set (n = 45) covering a range of clinical prescription scenarios. Plan quality was assessed by comparing automated plans against manual plans, with emphasis on target coverage and organ-at-risk (OAR) sparing. Three experts performed a blinded evaluation on the external validation set.
Figure 1 Superficial (top) and VMAT (bottom) dose calculations for nasal bridge basal cell carcinoma, with eye volumes contoured. Conclusion: The generation of a handbook of 3D Monte Carlo dose calculations for sample superficial radiotherapy treatments aids in staff training and provides valuable guidance around treatment options, which would not otherwise be available through the standard practice of manually calculating superficial point doses. References: [1] Nikandrovs M, McClean B, Shields L, McCavana P, Vintró LL. Clinical treatment planning for kilovoltage radiotherapy using EGSnrc and Python. J Appl Clin Med Phys 2023;24(2):e13832. https://doi.org/10.1002/acm2.13832. Keywords: superficial, kilovoltage, Monte Carlo
Made with FlippingBook - Share PDF online