ESTRO 2025 - Abstract Book

S276

Brachytherapy - Gynaecology

ESTRO 2025

4352

Digital Poster The impact of AI driven templates in brachytherapy practice

Maria Țîmpea 1,2 , Alexandru V Zariosu 1,2 , Sabina E Sucuri 3,2 , Diana V Cipu 4 , Simona Lupu 4 , Alina Munteanu 4 , Irina M Dumitru 4 , Cristina Sirbu 4 , Radu I Mitrica 1 , Amalia Constantinescu 1 , Xenia Bacinschi 1 , Marius Stanescu 5 , Lucian Bicsi 5 , Dragos Duse 5 , Dragos Grama 5 , Remus C Stoica 6,2 1 Radiation Oncology, Bucharest Institute of Oncology Prof. Dr. Alexandru Trestioreanu, Bucharest, Romania. 2 Medical Department, Synaptiq, Cluj-Napoca, Romania. 3 Radiation Oncology, Coltea Clinical Hospital, Bucharest, Romania. 4 Department of Medical Physics, Bucharest Institute of Oncology Prof. Dr. Alexandru Trestioreanu, Bucharest, Romania. 5 Research Department, Synaptiq, Cluj-Napoca, Romania. 6 Radiation Oncology, Global Medical Health, Bucharest, Romania Purpose/Objective: Brachytherapy, a cornerstone in radiotherapy, relies on precise dose planning to ensure effective tumor control while sparing healthy tissue. Advances in artificial intelligence (AI) have introduced automated contouring tools to delineate target volumes and critical structures, potentially revolutionizing the planning process. This study investigates the dosimetric differences between brachytherapy plans based on AI-contoured structures and those generated through manual contouring by experienced clinicians. This study advocates for a hybrid approach that combines AI’s efficiency with expert validation to optimize planning accuracy and ensure patient safety. Material/Methods: A retrospective analysis was conducted on nine patients undergoing gynecological brachytherapy either with vaginal cylinders or tandem-ovoid applicators, using pelvic CT scans for treatment planning. For each patient, OARs (bladder, rectum, and bowels) were contoured manually by clinicians and independently using the Mediq AI neuronal networks from Synaptiq. Dosimetric parameters, including D0.1cc and D2cc, were compared between the manually and AI-contoured OARs.

Results:

The analysis compared dose-volume histograms (DVHs) and D2cc values for rectum, bladder, and bowel between manual and AI-driven contouring methods. Regarding the time needed for adjustments of the AI contours, a mean of 224 seconds [104-484] was obtained. For the rectum, D2cc values ranged from 3.3Gy to 4.7Gy, with a mean of 4Gy for manual contouring, compared to a mean of 3.8Gy for AI segmentation. For the bladder, manual D2cc values ranged from 2.4Gy to 5.4Gy, with a mean of 4.5Gy, while the AI-contoured bladder showed a mean D2cc of 4.2Gy. The bowel, which showed the greatest variability in dose (ranging from 2Gy to 5Gy), had a mean D2cc of 3.2Gy for manual contouring, compared to 2.9Gy for AI segmentation. These findings, as shown in Figure1, suggest that AI driven segmentation may offer improvements in OAR sparing, particularly for the rectum and bowel, while maintaining comparable outcomes for the bladder.