Session Item

In March 2020, we clinically introduced ‘BRachytherapy via artificially Intelligent GOMEA-Heuristic based Treatment planning’ (BRIGHT) for prostate cancer patients. The intention behind BRIGHT is to overcome a time-consuming and unintuitive planning process, by automatically creating a set of high-quality treatment plans from which the physician can choose the preferred plan per patient. The purpose of this study is to evaluate the first clinical experiences with Artificial Intelligence (AI) based planning for HDR prostate plans.

Between March 2020 and January 2021, 7 prostate cancer patients were treated in our centre with single-dose HDR brachytherapy (BT) with a dose of 15 Gy. After implantation, MRI acquisition, catheter reconstruction, and delineation of the target volumes and organs at risk, BRIGHT was used for treatment planning (TP) (Fig. 1). BRIGHT consists of a bi-objective planning model, in which the TP aims (Table 1) are grouped into one coverage objective and one sparing objective, referred to as the Least Coverage Index (LCI) and Least Sparing Index (LSI). These were constructed in a worst-case manner: For example, an LCI = 2.0% means that all targets are covered at least 2.0% more than their planning-aim. BRIGHT’s graphical user interface allows navigation over the coverage-sparing trade-off curve and allows to make a selection for further dose-calculation and clinical plan determination in Oncentra Brachy. For each patient the plan selecting aspects and process times were monitored (Table 1).

For all patients, multiple plans that satisfied all planning aims could be selected in BRIGHT. Table 1 summarizes the results of the monitored parameters, reached planning aims, and LCI/LSI of the resulting clinical treatment plan for each patient. Time was spent on manual optimization of the BRIGHT plan before clinical plan approval to reduce large dwell times that caused undesired large high-dose sub-volumes in the target. Parameters of influence on the clinical plan choice were: Patient-specific clinical information, knowledge of the trade-off curve, dwell time pattern, and isodose distribution. As example: Based on these parameters the focus in patient 4 was on sparing of the urethra due to urinary complaints and the decision was made to underdose the target. In patient 2 a catheter was found incorrectly reconstructed, after correction the plan was not re-optimized in BRIGHT due to time pressure, resulting in a higher prostate V100%.

The users appreciated the intuitive process of navigating the trade-off curve to select the best solution for a patient. To improve time efficiency user training is on-going and additional planning criteria need to be implemented to improve plan quality, e.g. dose homogeneity and normal tissue dose constraints. After one year of clinical experience, we can conclude that AI based TP for HDR prostate BT is successfully implemented in our clinic and is a promising tool for further applications in BT.