ESTRO 2025 - Abstract Book

S2790

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

ESTRO 2025

The model demonstrated satisfactory accuracy, with Dice scores of 0.87 for 20Gy and 0.86 for 36Gy isodoses (Figure 1). A strong correlation (Spearman’s 0.8) was observed between MHD and H_V20Gy_plan. V20Gy_thr<4% was set, corresponding to MHD<3Gy. In Figure 2, H_V20Gy_plan and H_V20Gy_pred are plotted and clustered in quadrants according to the V20Gy_thr value. Sensitivity in identifying patients at higher risk was 90%, and the negative predictive value was 95%. Conclusion: This U-Net model accurately predicts MHD derived by TF planning, making feasible its use as an (immediate) decision tool, based only on the CT. A relevant application could be the possibility to robustly select patients that may benefit from breath-hold. Further validation on larger cohorts is recommended. Acknowledgments: Current study was supported by Italian Association for Cancer Research (AIRC)–IG23150.

Keywords: Breast cancer, AI isodose prediction, heart dose

References: [1] Stick LB, Lorenzen EL, Yates ES, Anandadas C, Andersen K, Aristei C, et al. Selection criteria for early breast cancer patients in the DBCG proton trial - The randomised phase III trial strategy. Clin Transl Radiat Oncol. 2021 Feb 4;27:126 131. doi: 10.1016/j.ctro.2021.01.012. PMID: 33659716; PMCID: PMC7892790.

1846

Poster Discussion Investigation of treatment plan complexity metrics to determine the quality of Radiotherapy Treatment Planning practice. Panagiota Aresti 1,2 , Brian Langan 2 , Margaret Moore 2 1 Physics Department, University of Galway, Galway, Ireland. 2 Dept Medical Physics & Clinical Engineering, University Hospital Galway, Galway, Ireland Purpose/Objective: Treatment plan quality is defined as the practical expectation of how well the delivered dose distribution matches the intended treatment plan and can be measured through qualitative and/or quantitative methods such as complexity metrics [1]. Although, numerous complexity metrics have been suggested, there is still no agreement on a standard approach [2]. A study conducted by ESTRO in 2020 highlighted the need for international guidelines for managing plan complexity, along with the consensus on the most suitable metrics [3]. This study aimed to: • Analyse eight complexity metrics derived from a commercial software. • Develop a framework that integrates these metrics for evaluating plan quality and optimising patient-specific quality assurance (PSQA). Material/Methods: A total of 140 treatment plans were retrospectively collected from four treatment sites. An independent treatment plan calculation software, called myQA® iON (IBA Dosimetry GmbH, Schwarzenbruck, Germany), calculated the following eight plan complexity scores (PCS) per treatment plan: field irregularity, leaf travel variability, leaf gantry synchronisation, small field contribution, off axis contribution, max leaf travel per monitor unit (MU), max gantry rotation per MU and beam delivered energy. The PCS were extracted and analysed, using the interquartile range (IQR) method to identify outliers. A “class solution” includes the acceptable range for each of the PCS for that treatment site. This study used the concept of “an outlier” to spot plans that deviated from the “class solution”.