S37
Brachytherapy - General brachytherapy
ESTRO 2026
stable and effective than image-based methods (Figure 1). SimpleElastix-BSpline achieved the highest prostate contour Dice score (0.94 ± 0.02), while SimpleElastix-Affine produced the lowest Hausdorff distance for biopsy contours (4.87 mm ± 1.58 mm). The OpenTPS-rigid pipeline offered comparable Dice and Hausdorff performance, while delivering the fastest computation time (0.89 s ± 0.01 s). Comparing RapidBrachyTG43 and RapidBrachyMC dose maps showed that all TG43-based DVH metrics were systematically overestimated by 3.33% to 6.82% (Figure 2). Among optimization solvers, Gurobi handled linear, quadratic, and hot spot penalty terms with the shortest solve time, whereas other packages were slower or failed on complex penalty function.Figure 1)
Mini-Oral 1923 BrachyUtils: A Python Platform for Benchmarking Registration, Dose, and Optimization in HDR Brachytherapy Hossein Jafarzadeh 1 , Jonathan Kalinowski 1 , Sébastien Quetin 1 , Farhood Farahnak 1 , Shirin A. Enger 1,2 1 Medical Physics Unit, McGill University, Montreal, Canada. 2 Lady Davis Institute, Jewish General Hospital, Montreal, Canada Purpose/Objective: High dose rate (HDR) brachytherapy improves long- term survival in prostate and cervical cancers. But due to its complex procedure, sub-optimal plans may be delivered to meet operational time constraints [1]. Although many automation tools have been proposed to streamline planning, consistent benchmarking of registration, dose calculation, and optimization methods on shared datasets and hardware is lacking. To address this, we developed BrachyUtils: a flexible Python framework that integrates in-house and open- source modules. Designed for scripting, it enables automated loops of patient-specific analyses for reproducible workflows and robust quality assurance in HDR brachytherapy. Material/Methods: BrachyUtils improves research efficiency through three design principles: leveraging open-source tools, adopting a modular architecture for flexible integration, and ensuring consistent and reproducible deployment across computing environments using Docker containers. It interfaces with three registration packages (OpenTPS [2], Plastimatch, SimpleElastix), two dose generation engines (RapidBrachyTG43 [3], RapidBrachyMC), and three optimizers (Gurobi, AMPL, ORTools). Docker Compose orchestrated isolated containers connected through a shared virtual network, enabling seamless inter-container communication. Registration algorithms were benchmarked using prostate biopsy data from µ- RegPro challenge [4], comparing image and contour- based registration between MR and transrectal ultrasound scans. Quality was assessed using the Dice similarity coefficient and Hausdorff distance. Dose generation and optimization modules were evaluated on 13 in-house HDR plans using CT or MR imaging. Dose generation was compared using DVH metrics and local and global percent error maps, following AAPM-WGDCAB Report 372. Solve time and penalty function complexity were used to benchmark optimization solvers. Results: BrachyUtils streamlines deployment through Docker containers and provides a unified Python interface for automated, multi-patient analysis. In registration evaluations, contour-based methods proved more
Figure 2)
Conclusion: By streamlining the integration of diverse planning tools and enabling high-throughput automated analysis, BrachyUtils simplifies research workflows in HDR brachytherapy and accelerates the development and validation of new approaches to treatment
Made with FlippingBook - Share PDF online