S141
Brachytherapy - Physics
ESTRO 2026
1 Radiation Oncology department, UZ Leuven, Leuven, Belgium. 2 Oncology department, KU Leuven, Leuven, Belgium Purpose/Objective: Today customized or 3D-printed skin applicators are commonly used in HDR brachytherapy. To fully exploit the advantages of these patient-specific devices, accurate placement on the patient should be ensured during each treatment fraction. This study investigates the potential of using surface guidance to verify the applicator position on the patient. Material/Methods: A surface scanning system was developed using a single depth-sensing Intel RealSense D405 camera, acquiring depth images through stereoscopic dual- RGB imaging at 30 fps. The surface scanning system was evaluated on an anthropomorphic facial phantom and a 3D-printed applicator (PolyJet 3D-printing; material: Stratasys Agilus30) to treat skin cancer in the nasal area (Figure 1). Per acquisition, 60 RGB and 60 depth images of the applicator surface and patient skin were captured and post-processed into one high- quality low-noise RGB and depth image, corrected for dead pixels.Based on RGB information, the applicator and facial phantom were automatically segmented in the depth images. For both applicator and the uncovered part of the face, a spatial point cloud was generated from the depth image, allowing the determination of their relative position in 6 degrees of freedom. To verify if the surface guidance method can detect applicator displacements, seven different setups were created with varying applicator positions (range up to 5.8 mm for translations and 6.2 ° for rotations). Each setup was surface captured ten times (to assess variability due to depth image fluctuations). Also a CT-scan (Siemens Somatom Edge, 1.5 mm slice thickness) was acquired for each setup from which a reference for the applicator position was determined based on rigid registration of the phantom and applicator. The relative applicator position with respect to the face phantom in each setup was compared between the two imaging methods.
Conclusion: We presented an end-to-end AI-assisted workflow for MRI-guided cervical HDR brachytherapy. We observed time savings with DVI differences attributed to variation in delineation. This workflow offers potentially faster AI-supported brachytherapy treatment planning with expert oversight. Keywords: delineation, reconstruction, optimization References: [1] F. Isensee et al. nnUNet: a self-configuring method for deep learning-based biomedical image segmentation, Nature Methods 18 (2021) 203–211.[2] V. Kostoulas et al. Dealing with segmentation errors in needle reconstruction for MRI-guided brachytherapy, Medical Imaging 2025: Image-Guided Procedures, Robotic Interventions, and Modeling, SPIE (2025) 529– 537.[3] D. Barten et al. Artificial Intelligence based planning of HDR prostate brachytherapy: first clinical experience, Radiotherapy and Oncology 161 (2021) S653–S655.[4] R. Pötter et al. The EMBRACE II study: The outcome and prospect of two decades of evolution within the GEC-ESTRO GYN working group and the EMBRACE studies, Clinical and Translational Radiation Oncology 9 (2018) 48–60. Digital Poster Highlight 4250 Surface guided brachytherapy for customized skin applicators Seppe Vanoppen 1 , Lore Sterckx 1 , Marisol De Brabandere 1 , Bertrand Dewit 2 , Tom Depuydt 1,2
Results:
Made with FlippingBook - Share PDF online