S2056
Physics - Image acquisition and processing
ESTRO 2026
Technology, Babe ș -Bolyai University, Cluj-Napoca, Romania. 10 Radiotherapy Department, ncology Institute “Prof. Dr. Ion Chiricuta”, Cluj-Napoca, Romania. 11 Department of Radiation Oncology, Hospital Clínic de Barcelona, Barcelona, Spain Purpose/Objective: To support QUARTET Euro-Ewing workshops1,2 testing delineation practices for Ewing sarcoma of two anatomical sites (pelvis and chest walls), we developed and validated an open-source workflow that converts multiple radiotherapy structure set (RTSTRUCT) contours into a voxel-wise Frequency-of-Inclusion (FoI) map and exports it as an RTDOSE suitable for visualization and percent-isodose consensus contouring in standard Digital Imaging and Communications in Medicine Radiotherapy (DICOM- RT) viewers. Material/Methods: A Python pipeline3,4 (pydicom, numpy, scikit-image) was developed to load an RT-Structure object and its associated DICOM CT series, selecting structures by keyword with optional exclusions. The script rasterizes the contours with slice assignment along the slice normal and builds binary masks indicating pixels belonging to the selected structures. A cumulative FoI map is generated on a user-selectable in-plane grid while preserving the CT field of view. For compatibility with standard clinical viewers, the FoI is exported within an RT-Dose object using a display mapping of one cGy per count5,6. Validation included a synthetic phantom and anonymised data from the QUARTET– Euro-Ewing multi-observer workshops. Results: On the phantom, script-derived isostructures closely matched computed with software algebra structures (union/overlap/intersection), demonstrating high geometric fidelity (Dice Similarity Coefficient 0.96-0.99, Average Surface Distance 0.14-0.30 mm, and maximum Hausdorff distance ≈ 1.65 mm) consistent with sub-voxel agreement. On anonymized dataset from the QUARTET Euro-Ewing multi-observer workshops (pelvis and chest wall cases), FoI heat maps localized unanimous cores, high-agreement shells, and low-agreement fringes; percent-threshold isostructures were generated to facilitate numerical analysis and to focus group review (Figure 1). Detailed clinical metrics and consensus outcomes will be reported separately.
Conclusion: Deep learning–based CBCT-to-CT synthesis generates sCTs with image and dose accuracy close to planning CT for lung SBRT, with SwinUNETR showing slightly more stable results. Further studies are needed to confirm these findings, investigate potential limitations, and assess robustness across different datasets and imaging conditions. References: 1. Thummerer A, van der Bijl E, Galapon A, et al. SynthRAD2025 Grand Challenge dataset: Generating synthetic CTs for radiotherapy from head to abdomen. Med Phys. 2025; 52:e17981. https://doi.org/10.1002/mp.17981 Keywords: adaptive radiotherapy, synthetic CT, deep learning Consensus by Design: Validating Frequency of Inclusion Heat Maps for Contour Agreement Analysis Volha Hertsyk 1 , Sarah M Kelly 2,3 , Melissa Christiaens 4 , Enrico Clementel 1 , Coreen Corning 1 , Raquel Dávila 5,6 , Jacob Engellau 7 , Maria Chiara Lo Greco 2,3 , Valentine Martin 8 , Andrada Turcas 9,10 , Jordi Saez 11 1 Headquarters, EORTC, Brussels, Belgium. 2 QUARTET Project, European Society for Paediatric Oncology (SIOP Europe), Brussels, Belgium. 3 Faculty of Medicine Digital Poster Highlight 1348 and Health Sciences, University of Ghent, Ghent, Belgium. 4 Department of Radiation Oncology, University Hospitals Leuven, Leuven, Belgium. 5 b. Department of Radiation Oncology, University Medical Center, Utrecht, Netherlands. 6 Pediatric Oncology, Princess Máxima Center, Utrecht, Netherlands. 7 Department of Radiation Oncology, Skane University Hospital, Lund, Sweden. 8 Département d'Oncologie- Radiothérapie, Gustave Roussy, Villejuif, France. 9 Institute of Advanced Studies in Science and
Figure 1 Pelvic case (five-observers subset). FoI colour- wash over representative CT planes showing agreement levels: red = 5/5, yellow = 4/5, green = 3/5, cyan = 2/5,
Made with FlippingBook - Share PDF online