S1527
Interdisciplinary - Quality assurance and risk management
ESTRO 2026
tracking and has demonstrated feasibility in early clinical use. The 1D interstitial and 3D rectal probe prototypes showed high mechanical robustness, precise geometric reproducibility, and dose linearity (R² > 0.99) during bench evaluation. All configurations withstood standard clinical sterilization processes. Clinical integration of the urinary catheter array added < 5 minutes to workflow and no adverse events were reported.
constant review by clinicians before clinical use, and in the event of major updates the tool must be revalidated.The aim of this work is to develop an easily-maintainable and scalable QA platform that automatically tracks contour editing over time, as a tool for continuous quantitative evaluation of autocontouring performances. Material/Methods: In our clinical practice, autocontours are generated using an AI-based tool (MIMProtegeAI v.1.3.2) external to our TPS (Eclipse Varian Inc). Automatically generated structures are reviewed and edited by radiation oncologists, who finally approve them for treatment planning.
Conclusion: By integrating 1D and 3D DOSEmapper™
configurations into routine applicators, the system enables practical and accurate in-situ dose verification across multiple treatment sites. This integrated engineering–clinical approach supports future multi- centre validation and the advancement of comprehensive treatment verification in brachytherapy. Keywords: Internal invivo dosimetry, Brachytherapy, QC
Figure 1 shows the workflow. A Python pipeline automatically retrieved patient data from the TPS. Anonymized images and contours were stored in an open-source PACS (Orthanc)[2]. Using an in-house library, DICOM data were processed to compute Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD) using MONAI[3]. Metrics were exported to a SQL- based database and visualized in Grafana[4], showing their evolution over time, stratified by clinician for each Organ-At-Risk (OAR).As a preliminary evaluation of platform performances, 66 prostate cancer patients treated between February and July 2025 were considered.For selected OARs, median and standard deviation were computed and compared against prior validation[5]. Statistical differences over time were assessed using the Mann-Whitney U test. Results:
Poster Discussion 3004
Development of a Scalable and Maintainable QA Platform for Monitoring AI-Based Autocontouring in Radiotherapy Gabriele Palazzo, Marco Paganelli, Sara Broggi, Roberta Castriconi, Antonella del Vecchio, Claudio Fiorino Medical Physics, IRCCS San Raffaele Scientific Institute, Milan, Italy Purpose/Objective: AI-based autocontouring has entered clinical practice in radiotherapy departments, resulting in time savings and reduced inter-observer variability (IOV)[1]. Automatically generated contours must undergo
Made with FlippingBook - Share PDF online