ESTRO 2025 - Abstract Book

S2782

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

ESTRO 2025

Conclusion: Our study demonstrates the efficacy of an advanced SBO method in automating IMRT treatment planning. SBO generated plans adhere to and have the potential to improve upon current treatment standards, while effectively balancing tumor coverage and minimizing doses to organs at risk. This approach sets new benchmarks in treatment planning quality and enhances the scope and precision of RT.

Keywords: Hyperparameter Optimization

References: 1. Balandat M, Karrer B, Jiang DR, et al. BOTORCH: A framework for efficient MC Bayesian optimization. Adv Neural Inf Process Syst. 2020;2020-Dec. 2. Huang C, et al. Meta-optimization for automated RT treatment planning. Phys Med Biol. 2022;67(5):05501 3. Hvarfner C, Hellsten EO, Nardi L. Vanilla Bayesian optimization in high dimensions. Phys Med Biol. 2024;235:20793-20817 4. Maass K, Aravkin A, Kim M. A hyperparameter-tuning approach to automated inverse planning. Med Phys. 2022;49(5):3405-3415 5. Pötter R, Tanderup K, Kirisits C, et al. The EMBRACE II study outcome. Clin Transl Radiat Oncol. 2018;9:48-60 6. Wang Q, Wang R, Liu J, et al. High-dimensional automated RT treatment planning via Bayesian optimization. Med Phys.2023;50:3773-3787

1811

Digital Poster Minimizing patient specific QA workload for VMAT breast treatments

Sara Poeta, Younes Jourani, Akos Gulyban, Diana Garcia Rodriguez, Jennifer Dhont, Nick Reynaert Medical Physics department, Institut Jules Bordet, Brussels, Belgium

Purpose/Objective: Pre-treatment measurements for patient-specific quality assurance (PSQA) are time- and labor-intensive when treating thousands of patients annually with volumetric modulated arc therapy (VMAT). This study seeks to reduce PSQA workload for breast treatments by correlating gamma index analysis with modulation complexity score. Material/Methods: Between January 2022 and March 2024, 678 breast cancer patients received treatments using VMAT techniques, targeting areas such as breast, lymph nodes (LN), internal mammary chain (IMN), and integrated boost (SIB). Treatment prescriptions varied from 26Gy to 40Gy/45Gy/48Gy in 5 to 15 fractions, respectively. Planning templates employed butterfly VMAT arcs (2TG) or 2 half arcs (HA), depending on treatment intent. Delta4 Phantom+ was used for QA on four mirrored Elekta Linacs. QAs were performed without daily correction factors, except for temperature adjustments. The initial gamma evaluation criteria were set at 3%/3mm, with a passing rate above 95%, and a stricter 2%/2mm criteria was later analyzed. Modulation complexity scores (MCS) were calculated, and scatter plots were used to determine acceptable cut-offs for QA exclusions. Results: Out of 678 breast cancer treatments, 33% were archived, leaving 453 plans in the study. All QAs initially passed the 3%/3mm gamma criteria. For those failing the stricter 2%/2mm criteria, phantom position optimization was applied; if successful, the QAs were marked as "Passed" (blue spots). "Failed QAs" (red spots) did not pass after optimization or were unaffected by daily corrections. Data is organized by treatment intent, dose prescription, and technique (Table 1). All plans using butterfly technique passed. For HA technique and breast-only plans, there was not enough data or results were poor to reach