S2394
Physics - Quality assurance and auditing
ESTRO 2026
within such a class solution. Currently, this might result in a subjective judgement call during plan classification. This research evaluates the use of an unsupervised data clustering algorithm on treatment plan complexity metrics2,3 for refining the definition
deviation of 0.1 mm Cu from manufacturer-reported values. Reference output measurements were within 2% of commissioning report data. The average difference between alanine-measured and treatment planning system-calculated dose across test cases of varying complexity was 6.05% ± 3.41%. Absolute average positional error determined using Gafchromic film was 0.22 mm ± 0.49 mm Conclusion: This study established a dosimetric baseline and tolerance range for preclinical targeted irradiations across London CRUK centres. These findings support future interinstitutional comparisons and enhance the reliability and translatability of preclinical radiotherapy research. They also provide key dosimetric indicators for routine performance checks enabling early detection of deviations and ensuring sustained accuracy across facilities. References: [1] Verhaegen F, Butterworth KT, Chalmers AJ, et al., “Roadmap for precision preclinical x-ray radiation studies,” Phys Med Biol, vol. 68, no. 6, Mar. 2023[2] Hill MA, Silvestre Patallo I, Aitkenhead AH, et al. The importance of standardization and challenges of dosimetry in conventional preclinical radiation biology research. Br J Radiol. 2025:1–12. [3] Stern W, et al. (2023) Facilitating the Adoption of Non-Radioisotopic Technologies in the Research Community: Reproducibility and Comparability. [4] Silvestre I, Subiel A, Westhorpe A, et al., “Development and Implementation of an End-To-End Test for Absolute Dose Verification of Small Animal Preclinical- Irradiation-Research-Platforms,” Int J Radiat Oncol Biol Phys, vol. 107, no. 3, 2020. Keywords: preclinical dosimetry, standardization Refining class solutions: clustering-based evaluation of treatment plan complexity Daan Hoffmans, Michelle Oud, Nicky van Lobenstein, Duncan den Boer Radiation Oncology, Amsterdam University Medical Center, Amsterdam, Netherlands Purpose/Objective: In the field of radiotherapy, the verification of treatment plan quality, in terms of deliverability and dosimetric accuracy, plays a major role in the overall process of quality assurance. A challenging aspect herein is the classification of which plans do or do not need further evaluation. The concept of class solutions, i.e. classification of plans based on treatment site, is accepted as a method for such pre- selection1. However, there is no consensus on the amount of (patient-specific) variation that is allowed Proffered Paper 4326
of a class solution. Material/Methods:
Our database was queried for VMAT plans used for the treatment of prostate, rectum and breast cancer on our TrueBeam (Varian, a Siemens Healthineers Company) treatment devices between 2016 and 2023. From these plans we calculated the following plan complexity metrics: Aperture irregularity, Edge Metric, Mean Field Area, Circumference Per Area, Small Aperture Score (5 mm) and MU/Gy2,3 using in-house developed software (Matlab, R2021a, Mathworks). For visualization, we calculated the principal components of these metrics (RStudio, v2023.06.1). The complexity metrics were standardized and a density based spatial clustering algorithm for applications with noise (DBSCAN)4 was used to detect clusters for each treatment site using parameters minPts = 11 and ε = 2. DBSCAN identifies points as seed points of a cluster when they have at least minPts points within euclidean distance e. Points that do not fulfill the criterium to be a seed point, but that are within distance ε from a seed point, are labeled as a border point of that same cluster. Points that are more than ε distant from any cluster are identified as outlier4. Results: The first two principal components of the complexity metrics of all treatment sites are shown in Figure 1, where separation between the treatment sites is visible. The clusters and outliers are shown in Figure 2.
Made with FlippingBook - Share PDF online