S1592
Physics - Autosegmentation
ESTRO 2026
Oncology, Calvary Mater Newcastle, Newcastle, Australia. 12 Department of Radiation Oncology, Calvary Mater Newcastle, Newcastle, Australia Purpose/Objective: Radiation dose to the heart influences cardiac side effects. A number of cardiac side effect prognostic models have been developed that include dose- volume metrics to predict events1,2. Heart contouring variation affects these metrics and may alter modelled risk. We applied a validated autosegmentation model to compare whole-heart contours from automated and clinical sources across 8 services (14 hospitals), assessing dose-volume metric differences that could impact outcome modelling and future model development. Material/Methods: Six radiotherapy services involved in the Australian Cancer Data Network collected standardised datasets for both lung and breast cancer radiotherapy cohorts; one service for lung only and one for breast only. Patient numbers, cohort selection criteria and the factors recorded for each patient are as described in Table 1. At local services an open source cardiac autosegmentation model3 was used to segment the whole-heart on planning CTs. Contour and dose- volume metrics were computed for both autosegmented and clinical contours using a Python script. Cohort averages and standard deviations were calculated locally, then combined centrally from the local summary analysis. Heart contour differences were considered across centres and dose-volume metrics were analysed by anatomical site and
heart V20 differences ranged from 17 - 40% and similar to MHD (Fig 1c) and were reduced with VMAT and IMRT compared to 3D CRT.
Conclusion: The difference between using clinical contours and a consistent automated contour was similar for both breast and lung patients and across services. The difference in resulting dose-volume metrics was much larger for patients with lung cancer than breast cancer. Modelled cardiac side effects using dose-volume metrics for lung cancer patients should consider uncertainty due to contour variation and future models should incorporate this uncertainty. References: 1. van den Bogaard, V.A., et al., Validation and Modification of a Prediction Model for Acute Cardiac Events in Patients With Breast Cancer Treated With Radiotherapy Based on Three-Dimensional Dose Distributions to Cardiac Substructures. J Clin Oncol, 2017. 35: p. 1171-1178.10.1200/jco.2016.69.84802. Tohidinezhad, F., et al., Prediction models for treatment-induced cardiac toxicity in patients with non-small-cell lung cancer: A systematic review and meta-analysis. Clin Transl Radiat Oncol, 2022. 33: p. 134-144.10.1016/j.ctro.2022.02.0073. Finnegan, R.N., et al., Open-source, fully-automated hybrid cardiac substructure segmentation: development and optimisation. Physical and Engineering Sciences in Medicine, 2023. 46: p. 377-393 Keywords: cardiac dose, uncertainty, outcome
treatment approach.
Results: Contour differences were consistent across anatomical sites and services (Fig 1a). Mean Dice Similarity Coefficient (DSC) was greater than 0.9, Mean Distance to Agreement (MDA) was less than 3 mm, and volume differences were within 10–50 cc for all anatomical sites and across services. For left whole- breast Mean Heart Dose (MHD; Fig 1b) and heart V20 differences between clinical and automated contours were less than 1 Gy and 7% respectively and were reduced with use of deep inspiration breath hold (DIBH). Differences increased with nodal irradiation as well as whole-heart, especially internal mammary chain and were greater with VMAT than IMRT. Doses were reduced for right breast. For lung cancer, mean
Made with FlippingBook - Share PDF online