ESTRO 2025 - Abstract Book

S359

Brachytherapy - Physics

ESTRO 2025

Material/Methods: The Topas MC code was used to study the effectiveness of ocular lead shields for an interstitial brachytherapy of the nose. The Topas brachytherapy source package for 192-Ir TG-186 source was used. A python script was written to read DICOM-RT files (source position, rotation and dwell-times) of the patient (anonymized data). Topas Geant 4 option 4 with fluorescence on was used. Different cases of shielding were investigated; lead of 2mm, 6mm and 4mm lead with 2 mm copper (posterior to the lead) on the top of the left and right eye. Results: Preliminary results show that all shields tested lowered the dose to the eye compared to use of no shield and that switching from pure lead to include also copper coating could reduce the characteristic X-rays from lead. Further studies on optimal shield design by combination of materials are ongoing as is calculations for other facial implant locations. Conclusion: This work indicated that lead effectively shielded the eyes for the studied patient. However, combing lead with material of lower atomic number, like copper, to reduce characteristic x-rays from lead could be more effective than use of a pure lead shield.

Keywords: ocular shielding, facial implants, 192-Ir

References: Candela Juan et al, 2014 "Dosimetric perturbations of a lead shield for surface and interstitial high-dose-rate brachytherapy", J. Radiol. Prot. 34 297

4002

Digital Poster Evaluating the utility of AI-based automatic needle reconstruction for MRI-guided brachytherapy of cervical cancer Vangelis Kostoulas 1 , Lavinia Verhagen 1 , Ellen Kerkhof 1 , Peter A.N. Bosman 2 , Tanja Alderliesten 1 1 Radiotherapy, Leiden University Medical Center, Leiden, Netherlands. 2 Evolutionary Intelligence, Centrum Wiskunde & Informatica, Amsterdam, Netherlands Purpose/Objective: Numerous Automatic Needle Reconstruction (ANR) methods exist. Often, they are only evaluated by investigating deviations from Clinical Needle Reconstructions (CNR). This study assesses the utility of AI-based ANR in MRI-guided cervical cancer brachytherapy, focusing on observer agreement and qualitative insights for clinical use. Material/Methods: We included 108 patients with corresponding T2-weighted and BFFE MRI scan pairs acquired with Philips Ingenia 1.5T/3T MRI systems, alongside clinical needle annotations of tip and bottom points. Patients were treated using an Utrecht or Venezia applicator with on average 4 needles per patient (ProGuide 6F Round, Elekta). The dataset was split into three subsets, each subset serving as a test set while the remaining data was used to train a 3D nnUNet [1]. The dataset was then grouped again, into two subsets: those for which the test segmentations had errors and those for which they did not. From either subset, 10 patients were chosen at random. One patient was excluded due to inconsistencies between the T2 and BFFE images, leaving a total of 19 patients. The final set included 84 needles.