ESTRO 2024 - Abstract Book

S345

Beachytherapy - Physics

ESTRO 2024

An AI-based bi-objective optimization approach for prostate HDR brachytherapy treatment planning, named BRIGHT, has been clinically introduced at the Amsterdam University Medical Centers in 2020 [1]. BRIGHT optimizes directly on existing dose volume indices (DVIs) as described in a clinical protocol and results in a set of treatment plans ranging from high coverage of target volumes to high sparing of organs at risk (OARs). Whilst this approach produces high-quality plans, existing comorbidity like Crohn’s disease could result in the preference to specifically further spare the sigmoid or rectum. We therefore developed and evaluated an extension for the existing approach to automatically find additional sets of treatment plans that each focus on sparing a specific OAR.

Material/Methods:

The existing optimization model has two objectives [2], the Least Coverage Index and Least Sparing Index. Either objective consists of an exponentially weighted ordered sum of the DVIs pertaining to coverage or sparing aims in a clinical protocol, respectively. At any time during optimization, the most violated or least satisfied DVI is predominantly optimized on by assigning it the largest weight. The AI technique used for optimization is a version of the Gene-pool Optimal Mixing Evolutionary Algorithm [3] that incorporates a GPU-accelerated dose calculation engine [4] to compute the DVIs according to the AAPM TG43 model [5] on 100.000 dose calculation points (DCPs). Results are re-evaluated on 500.000 DCPs. The proposed extension optimizes not a single set of trade-off solutions, but multiple sets, one for each OAR and one for solely the clinical protocol without any additional OAR sparing. Each OAR sparing set is optimized to spare their appointed OAR more in comparison to plans from the other sets, without necessarily violating the clinical protocol aims. This is achieved by incorporating a third objective, called “added OAR sparing”, that is to be maximized for each plan p . See Figure 1A for a visualization. The objective value for p is the minimum difference of its DVI of the appointed OAR of the set that p is in, and the same DVI of plans from the other sets s as follows: As we introduce an additional objective, all plans are still optimized on the existing sparing versus coverage trade off to adhere to the clinical protocol. For prostate brachytherapy, the outcome consists of 4 different sets of treatment plans: one set pertaining to only the clinical protocol, and three sets pertaining to additional sparing of the bladder, rectum, and urethra. All sets of treatment plans are obtained from a single run of the approach to mimic clinical use. For evaluation, treatment plans were retrospectively optimized with the extended approach for 20 patients. From those, 5 patients with and 5 patients without all clinical aims satisfied were randomly selected to provide a representative sample. For each patient, the plan that equally balances the coverage and sparing objectives was selected from the protocol only set. From the sets of treatment plans that focus on the bladder, rectum, and urethra, at least one, but possibly more plans were selected. A medical physicist and radiation oncologist were then presented independently with the selected plans. Both observers were blinded for the sets in which the plans belong. They were asked to correctly categorize the presented plans into each of the four sets to determine if there is a focus on the additional sparing of a specific OAR, and, if this is the case, which OAR.

Results:

The numbers of correctly categorized treatment plans are given in Table 1. For illustration, Figures 1B and 1C display two presented treatment plans from different sets. For increased sparing of a specific OAR (see blue points in Figure 1A), in on average 93% (89/96) of the cases the correct categorization was identified. Treatment plans from the protocol only set were miscategorized mostly as plans from the set optimized for the urethra. The distinction between these can be difficult as the urethra, centrally located in the prostate, is usually a limiting factor if a plan with equal balance between coverage and sparing is chosen. No meaningful differences were observed between patients with and without all clinical aims satisfied.