ESTRO 2025 - Abstract Book

S3590

Physics - Quality assurance and auditing

ESTRO 2025

1 Department of Radiation Oncology, Netherlands Cancer Institute, Amsterdam, Netherlands. 2 Department of Radiation Oncology, Radboud University Medical Center, Nijmegen, Netherlands. 3 Department of Radiation Oncology, UZ Leuven, Leuven, Belgium. 4 Department of Oncology, KU Leuven, Leuven, Belgium. 5 Department of Radiation Oncology, AZ Sint-Maarten, Mechelen, Belgium. 6 Department of Radiation Oncology, Radiotherapiegroep, Arnhem/Deventer, Netherlands. 7 Department of Radiation Oncology, Radiotherapeutic Institute Friesland, Leeuwarden, Netherlands. 8 Department of Radiation Oncology, AZ Turnhout, Turnhout, Belgium Purpose/Objective: Focal boosting in radiotherapy of prostate cancer comes with two challenges that may limit wider implementation. Firstly, inadequate GTV contours may lead to inferior outcome [1]. Secondly, reaching a sufficiently high tumor dose can be challenging in the proximity of organs at risk and depends to some extent on the experience of the planner. To ensure quality in a prospective multicenter trial, we implemented centralized prospective individual case review. In this work, we describe the development and performance of an automated QA system and its application on the experimental arm of the multi-center hypo-FLAME 3.0 trial on focal boosting in prostate cancer. Material/Methods: We used a secure online cloud storage service (Surfdrive) to facilitate DICOM data transfer. We developed a system in Python that does preprocessing (consisting of standardizing anatomical structure names and scan descriptions, CT to MRI registration), applies evaluation models and generates reports with evaluation results, which are automatically sent by e-mail to the central reviewers. Based on this email, they advise the participating center to continue with the treatment or to reconsider the segmentation and/or treatment plan. To evaluate GTV contouring, we applied a model published earlier for the evaluation of the FLAME trial [2]. Assessment by the reviewers was based on both the model output and the visualisations of multiparametric-MRI at the tumor location. To evaluate plan quality, we developed and implemented a knowledge based random forest regressor model to predict the near minimum dose (D98%) to the GTV, based on anatomical features of the GTV. Assessment by the reviewers was based on both the model output and DVH statistics. Results: In the study arm 68 patients were included. The response times to external institutes and QA results of all patients in the study arm are listed in Table 1. Contouring quality is reported per patient, whereas planning quality is reported per GTV. In 2 cases, recontouring was carried out and after uploading a new case, the model performance improved. An overview of the planning QA results is shown in Figure 1. In 6 cases, a replanning was recommended resulting in a higher D98% to the GTV 4 times. In 3 cases, the planned D98% was less than the tolerance, but plan QA was still passed due to DVH statistics considerations.