Session Item

In 2020, ‘BRachytherapy via artificially Intelligent GOMEA-Heuristic based Treatment planning’ (BRIGHT) for prostate HDR BT was clinically introduced. BRIGHT is a bi-objective treatment planning method that finds a set of high-quality, patient-specific treatment plans (TPs) with different trade-offs between clinical target coverage and organ sparing, by directly optimizing the dose volume indices (DVIs) in the clinical protocol. However, in the clinic, manual adjustments of BRIGHT TPs are still done to meet additional patient specific aims. Particularly, this includes minimization of contiguous high-dose volumes, i.e., hotspots (HSs). We therefore aim to incorporate control of HS volumes in BRIGHT, while minimally impacting obtainable DVI values.

We augment BRIGHT with a third objective to minimize HSs. For this, we define an HS as ‘a contiguous volume of >0.1 mL outside catheters receiving >300% in target volumes: prostate and seminal vesicles, or >200% in normal-tissue around target volumes of the prescribed dose’. We tailored a graph-based method, which uses a connected component algorithm (Afforest), to determine HSs. The graph consists of dose calculation points (DCPs) as nodes and edges between close (≤0.5 mm) neighbouring DCPs. DCPs are randomly sampled locations where the dose is calculated (to compute the DVIs).

The third objective in tri-objective BRIGHT is the sum of HS volumes (metric 1). For comparison, we also consider as third objective a more efficiently computable metric, which however ignores whether the high-dose volume is contiguous: the sum of V_300% of the target volumes and V_200% of normal tissue (metric 2).

We compare bi-objective BRIGHT with both tri-objective BRIGHT versions on a data set of 11 prostate cancer patients by retrospectively planning single-dose HDR BT with 

Figure 1 shows for patient 9 that, both tri-objective BRIGHT versions result in a clear improvement in control of HSs; TPs with HSs ≤0.5 mL are only found using the tri-objective versions. Due to the nature of metric 2 and using a 2D plot for a 3D front, the trade-off between existing DVIs and metric 2 culminates in a larger covered area with TPs with low HS volumes.

Table 1 shows that when using metric 1, for 10 out of 11 patients, total HS volume could be reduced to ≤0.5 mL while satisfying the clinical protocol, versus 6 out of 11 patients for bi-objective BRIGHT. Adding metric 1 does not result in worsening of DVIs. Adding metric 2 does cause slight worsening of DVIs but results in more plans satisfying the clinical protocol without HSs.

Currently, using metric 1 and 2 takes 1800s and 600s, respectively. Metric 1 needs further optimization to definitively assess runtime impact.

We successfully adapted BRIGHT to reduce HSs without compromising obtainable DVI values for most patients, by explicitly computing HSs and minimizing their volume through a third objective. This could potentially render manual HS adjustments redundant.