ESTRO 2025 - Abstract Book

S3506

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTRO 2025

2677

Digital Poster Automated proton therapy treatment planning pipeline for head and neck cancer patients Poppy Nikou 1,2,3 , Anna Thompson 4 , Teresa Guerrero Urbano 5,6 , Andrew Nisbet 1 , Jamie McClelland 1,2 , Sarah Gulliford 3,1 1 Medical Physics and Biomedical Engineering, University College London, London, United Kingdom. 2 UCL Hawkes Institute, University College London, London, United Kingdom. 3 Radiotherapy Physics, University College London Hospital, London, United Kingdom. 4 Radiotherapy, University College London Hospital, London, United Kingdom. 5 Radiotherapy, Guys and St Thomas' Hospital, London, United Kingdom. 6 Comprehensive Cancer Centre, Kings College London, London, United Kingdom Purpose/Objective: Proton therapy treatment planning is a complex optimisation problem and a time-intensive task. Knowledge-based treatment planning systems and scripting have been used to optimise this workflow [1-3]. However, these approaches do not automate the processing of the radiotherapy structure sets. This work addresses this limitation by developing and describing a pipeline for automated proton planning. Material/Methods: This study included 13 head and neck cancer patients with bilateral target volumes. The radiotherapy planning structures and proton plans were generated using Python scripts. Planning structures were created using anatomical reference points, defined through a rigid registration of the patient’s planning CT (pCT) to an atlas, as illustrated in Figure 1. The remaining growing and Boolean operations were performed using a script within the Raystation Python environment (Raysearch Labs, Stockholm/Sweden).

Plans were created in Raystation with two dose levels (65Gy, 54Gy in 30#). Two treatment plans were set-up: Plan A (0⁰, 45⁰, 180⁰, 315⁰) and Plan B (45⁰, 100⁰, 260⁰, 315⁰). A template of objectives and clinical goals was imported to optimise and evaluate the treatment plans. Robustness scenarios incorporating 3mm shifts and 3.5% range uncertainty were considered. Results: The pipeline generated a plan in ~20-30 minutes, with optimisation time varying according to the target volume size. Small manual adjustments were required for the spot target volumes, including deleting small islands/slices or smoothing. Figure 2 shows the nominal and worst-case scenario D 95 for the high and low dose CTV, D max for the brainstem and spinal cord, and D mean for the parotid glands for both plans. The worst-case analysis excluded the parotid glands. The high dose CTV median D 95 met the requirements in all plans, with one outlier resulting from conflicts between target coverage and parotid gland sparing during optimisation. The low dose CTV median D 95 met the requirements in the nominal case but was below (<0.2Gy) in Plan B’s worst-case scenario. The median D max for the brainstem and