S22
Brachytherapy - Gastro-intestinal, paediatric brachytherapy, miscellaneous
ESTRO 2026
and centroid distance (CD). Results:
217/28, 60,7%3/28, 10,7% 8/26 28,6%, Conclusion: After proposed treatment program major
Scripting-assisted preprocessing achieved high segmentation consistency, including CT datasets without O-MAR correction. Trachea contour segmentation yielded DSC = 0.99 ± 0.05 and Jaccard = 0.98 ± 0.07. Balloon segmentation achieved DSC = 0.89 ± 0.17 and Jaccard = 0.83 ± 0.17, with smooth surface representation for balloon volumes of 2–6 ml. For GTVs, the mean DSC exceeded 0.95, confirming robustness. A representative test case achieved DSC = 0.98 ± 0.02 and HD95 = 2.0 ± 1.0 mm. The workflow reduced inter-operator variability and computation time from approximately 45 to 17 minutes. Conclusion: The proposed intelligent scripting workflow enables accurate and efficient CT-bases segmentation for esophageal HDR brachytherapy, particularly when combined with transnasal balloon applicators to enhance clinical workflow efficiency. For cases without stent implantation, maintaining balloon applicator conformity was essential for achieving dose constraints (D90%) in treatment planning. For cases involving stent implantation(high density substance), based on EBT3 film and Lucite phantom verification, accurate dummy&catheter delineation (2.4 mm diameter replaced by air-cavity) and Monte Carlo– based AcurosBV dose calculation remained crucial for ensuring planning consistency (5.8–7.5% D90% dose variation). Future studies will integrate external beam radiotherapy data and prospective validation to assess clinical scalability. Keywords: esophageal, balloon applicator, auto- segmentation References: 1. Taggar AS, Pitter KL, Cohen GN, et al. Endoluminal high-dose-rate brachytherapy for locally recurrent or persistent esophageal cancer. Brachytherapy 2018;17:621–627 . 2. Bin Cai*, Michael B. Altman, et al. Standardization and automation of quality assurance for high-dose-rate brachytherapy planning with application programming interface : Brachytherapy 2019; 108:1143. Michelle Oud, ⁎ , Inger-Karine Kolkman- Deurloo, Jan-Willem Mens, Danny Lathouwers, Zoltán Perkó, Ben Heijmen, Sebastiaan Breedveld : Fast and fully-automated multi-criterial treatment planning for adaptive HDR brachytherapy for locally advanced cervical cancer: Radiotherapy and Oncology 148(2020) 143-150
(cCR+pCR+ncCR) response can be achifed in 90% T2N0-1 and in 71.4% T3N0-2 patients with low rectal cancer Keywords: rectal cancer, complete response, watch and wait
Digital Poster 2013
Intelligent CT Auto-Segmentation and Scripting Workflow for Enhanced HDR Brachytherapy Planning Accuracy in Esophageal Cancer Jian-Kuen Wu 1 , Jason Chia-Hsien Cheng 1 , Yen-Ting Liu 2 , Hsing-Lung Chao 3 , Feng-Ming Hsu 1 1 Division of Radiation Oncology, Departments of Oncology, National Taiwan University Hospital, Taipei, Taiwan. 2 Division of Radiation Oncology, Departments of Oncology, National Taiwan University Hospital, Yunlin Branch, Yunlin, Taiwan. 3 Department of Radiation Oncology, Wan Fang Hospital, Taipei, Taiwan Purpose/Objective: This study introduces and evaluates an intelligent workflow integrating CT auto-segmentation with scripting via an application programming interface (API) in a commercial treatment planning system (TPS). The workflow was developed for esophageal cancer patients treated with transnasal balloon applicators, addressing challenges associated with variable balloon volumes, stent implantation, and dummy source artifacts, with the goal of improving segmentation accuracy and enhancing overall clinical workflow efficiency. Material/Methods: A retrospective analysis was conducted on 22 patients (45 CT datasets, including 6 with stent implantation) treated with high-dose-rate (HDR) esophageal brachytherapy using the Varian Eclipse system. Each dataset included balloon-filled volumes of 2–6 ml, acquired between treatment weeks 1 and 3. The O- MAR algorithm was applied for artifact reduction and image renormalization. Tracheal contours served as internal quality-control benchmarks; only datasets achieving a Dice Similarity Coefficient (DSC) > 0.90 and Jaccard index > 0.85 were included for model training. Gross tumor volume (GTV), balloon curvature, dummy source (catheter reconstruction), and trachea were manually delineated by five clinicians to generate consensus contours. A U-Net–based model was trained using the holdout method (8 for training, 1 for validation, and 1 for testing). Performance was evaluated using DSC, 95th-percentile Hausdorff distance (HD95), average symmetric surface distance (ASSD), relative absolute volume difference (RAVD),
Made with FlippingBook - Share PDF online