ESTRO 2025 - Abstract Book

S3557

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

ESTRO 2025

References: [1] Sørensen et al. Proton FLASH: impact of dose rate and split dose on acute skin toxicity in a murine model. Int J Rad Oncol Phys 2024: 120; 265-275. [2] Poulsen et al. Oxygen enhancement ratio weighted dose quantitatively describes acute skin toxicity variations in mice after pencil beam scanning proton FLASH irradiation with changing doses and time structures. Int J Rad Oncol Phys 2024: 120; 276-286. [3] Kristensen et al. Spread-out Bragg peak FLASH proton therapy: quantifying tissue toxicity in a mouse model. Front Oncol 2024: 14, 1427667.

4117

Digital Poster A novel probabilistic methodology for optimizing under- and overdosage Jelte R de Jong, Danny Lathouwers, Zoltán Perkó Applied Sciences, Radiation, Science & Technology, Delft, Netherlands

Purpose/Objective: Treatment planning aims to achieve sufficient tumor coverage while sparing surrounding tissues, thereby maximizing the therapeutic window. Uncertainties are commonly handled by robust optimization [1], which uses a limited set of scenarios, optimizing for the worst-case. While effective, this may result in overly conservative plans or insufficient robustness if too few scenarios are considered. Probabilistic planning allows for the use of statistical measures [2,3]. We present a proof-of-principle probabilistic methodology enabling users to define probabilities for voxel-wise under/overdosage, optimized through percentile approximations by expectation value and standard deviation (SD). We compare this approach with a robust equivalent and demonstrate its potential of greater tissue sparing. Material/Methods: We consider a model geometry of dimensions 57x57x57 mm 3 with 3x3x3 mm 3 voxels, with a spherical CTV (prescribed 60Gy) and OAR (aiming for ≤45Gy) positioned along the 45-degree axis in the lateral-depth plane. Gaussian distributed systematic setup and range uncertainties (3mm/3%) are addressed by Polynomial Chaos Expansion [4] to generate dose scenarios. The probabilistic methodology optimizes for an approximate percentile in an inner loop, with an outer loop ensuring the approximation matches the user-defined probabilities. The dose distribution from a composite robust plan for CTV and OAR (5mm/5%) is scaled to match the D98 10th-percentile of the probabilistic plan, ensuring probabilistically identical target coverage. Results: Figure 1 shows the cumulative distribution functions (CDFs) for the D98 and D2 of the robust and probabilistic plans, scaled to ensure that 90% of the scenarios have a D98 above 51Gy. The robust plan increases CTV overdosage (D98 90th-percentile) by 5.4% compared to the probabilistic plan. The probabilistic plan spares more OAR in the high-dose region, achieving higher cumulative probabilities across the D2 range. Moreover, it achieves better CTV conformity for the 95%-isodose line, reducing the V95% of the geometry by 16.4%. Figure 2 compares the standard deviation volume histogram (SDVH, DVH made from voxel-wise dose standard deviation values) for both plans. The probabilistic plan reduces SD for the CTV and OAR, while maintaining similar SD for tissue, achieving lower dose variability under uncertainty.