ESTRO 2025 - Abstract Book

S2845

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2025

References: [1] Griffin RJ, Ahmed MM, Amendola B, Belyakov O, Bentzen SM, Butterworth KT, et al. Understanding High-Dose, Ultra-High Dose Rate, and Spatially Fractionated Radiation Therapy. Int J Radiat Oncol Biol Phys 2020;107:766– 78. https://doi.org/10.1016/j.ijrobp.2020.03.028. [2] Amendola BE, Perez NC, Mayr NA, Wu X, Amendola M. Spatially fractionated radiation therapy using lattice radiation in far-Advanced bulky cervical cancer: A clinical and molecular imaging and outcome study. Radiat Res 2020;194:724–36. https://doi.org/10.1667/RADE-20-00038.1. [3] Amendola BE, Perez NC, Wu X, Amendola MA, Qureshi IZ. Safety and Efficacy of Lattice Radiotherapy in Voluminous Non-small Cell Lung Cancer. Cureus 2019;i. https://doi.org/10.7759/cureus.4263.

2972

Proffered Paper GPT-RadPlan: a plugin for automated treatment planning in Eclipse TPS based on large language models Oscar Pastor-Serrano 1 , Sheng Liu 1 , Shirley Cheng 1 , Peng Dong 1 , Yong Yang 1 , Thomas Niedermayr 1 , Elizabeth Kidd 1 , James Zou 2,3 , Lei Xing 1 1 Department of Radiation Oncology, Stanford University, Stanford, USA. 2 Department of Biomedical Data Science, Stanford University, Stanford, USA. 3 Department of Computer Science, Stanford University, Stanford, USA Purpose/Objective: Radiotherapy treatment planning is a time-consuming and potentially subjective process that requires iterative adjustments of optimization parameters to balance multiple conflicting objectives. In this study, we aim at leveraging the in-context learning capabilities and prior radiation oncology knowledge encoded in large language models (LLMs) such as OpenAI's GPT-4Vision to automate treatment planning. The resulting GPT-RadPlan integrates directly with the clinically widely used Eclipse Treatment Planning System (TPS) and acts as an expert human planner capable of guiding the planning process by adjusting optimization weights, objective doses, and dynamically adding or removing planning objectives. Material/Methods: GPT-RadPlan iteratively evaluates treatment plans and suggests new optimization settings based on its own suggestions. At the beginning of each iteration, GPT-RadPlan’s evaluation module separately assesses dose distributions and dose-volume histograms (DVHs), providing detailed textual feedback on plan quality checking the level of agreement with the physician’s intent. Based on (i) the evaluation module’s feedback, (ii) the optimization settings of previous iterations, and (iii) the optimization settings from three previous approved plans for the same disease site and prescription; the planner module adjusts optimization weights, modifies objective doses, and adds or removes planning objectives to meet the clinical requirements. The efficacy of GPT-RadPlan was evaluated on 6 prostate cancer cases undergoing volumetric modulated arc therapy (VMAT). Results: GPT-RadPlan-generated plans were compared to clinically approved plans for 70.20 Gy three-arc VMAT prostate cancer treatments, obtained from human planners at our institution. In all cases, GPT-RadPlan either outperformed or matched the clinical plans, similarly covering the targets while reducing organ-at-risk (OAR) doses 15% on average. Specifically, GPT-RadPlan achieved an average bladder and rectum mean dose of 27.5 Gy and 31.08 Gy, compared to 29.55 Gy and 36.33 Gy for the clinical plans, respectively. The proposed method consistently satisfied all dosimetric objectives outlined in the physician’s intent after 3-6 planning iterations, demonstrating its ability to adjust planning parameters dynamically and optimize plan quality.