The versatility of evolutionary intelligent tri-objective treatment planning for cervical cancer brachytherapy

Med Phys 2025; 52(8). doi:10.1002/mp.18022

Leah R.M. Dickhoff, Ellen M. Kerkhof, Heloisa H. Deuzeman, Danique L.J. Barten, Laura A. Velema, Lukas J.A. Stalpers, Bradley R. Pieters, Carien L. Creutzberg, Peter A.N. Bosman, Tanja Alderliesten

What was your motivation for initiating this study?

Automated treatment planning has many potential benefits, not the least of which are better plan quality and workflow optimisation. However, the automation of cervical cancer brachytherapy planning is complex. Practices vary widely between countries or even institutions. Consequently, there is no single definition of a “best” plan – what is considered optimal at one centre may not be at another, depending on clinical philosophy, protocol interpretation, and patient-specific factors. Most automated planning approaches either optimise on mathematical surrogates rather than clinically used dose-volume metrics, or they generate only one plan, which then locks all clinicians and patients into a single definition of optimality.

This situation highlights the need for a versatile method that is not only aligned with established protocols but also can be easily customised to reflect institution-specific preferences. The previously developed and currently clinically used approach for prostate brachytherapy is called BRIGHT (brachytherapy via artificially intelligent GOMEA heuristic-based treatment planning, in which GOMEA refers to a family of evolutionary algorithms). It uses a bi-objective model that directly optimises tumour coverage and organs-at-risk (OAR) sparing, and that provides multiple alternative plans that trade-off coverage and sparing intuitively via a unique “always focus on improving the worst-case first” combination approach. BRIGHT optimises only the dwell times, using pre-defined catheter reconstructions, delineations, and prescribed doses. Our motivation was to extend BRIGHT from prostate to cervical cancer brachytherapy, and to ensure it could flexibly accommodate potential diverse clinical preferences.

What were the main challenges during the work? 

A key challenge was how to incorporate additional planning aims – to shape the dose distribution in a certain way, which is important for clinical acceptability but can also reflect institution-specific preferences – while still providing immediate insight into whether the aims of the EMBRACE-II protocol regarding tumour coverage and OAR sparing were met. To accomplish this, we introduced a third optimisation objective, which allowed optimisation on additional aims without compromising or obscuring the EMBRACE-II protocol requirements. This feature made the approach versatile, since new aims could easily be added to it while giving insight into the trade-off between the additional aims in the third objective and the EMBRACE-II aims. Another distinctive challenge was the inclusion of contiguous volume optimisation, which BRIGHT can uniquely take into account during optimisation. This allows optimisation for a single high-dose region, and thereby ensures contiguous dose distributions, which are often preferred in cervical brachytherapy.

What are the most important findings of your study?

We have found that the new tri-objective version of BRIGHT, when applied to intracavitary-interstitial cervical cancer brachytherapy cases, can generate plans in just a few minutes, even when using highly precise dose calculations during optimisation. BRIGHT’s flexibility allowed us to test different customisations, including explicitly pear-shaped distributions and generally-contiguous volume optimisation. Comparable dose-volume values could be achieved with all customisations, though with observable variations in the resulting dose distributions. Importantly, when evaluated by a brachytherapy medical team, at least one plan from each set was considered clinically acceptable for every patient reviewed.

What are the implications of this research?

BRIGHT not only produces plans that comply with EMBRACE-II protocol aims, but is also easily customisable and provides clinicians with meaningful choices among multiple high-quality options. This can help to reduce workload, improve plan quality, increase insight into possibilities, and broaden access to high-quality treatment. BRIGHT generates plans in under four minutes – far faster than manual planning and many existing systems – and additional speed-ups, many of which can lead to similar quality plans in less than 10-30 seconds, are still being researched. An offer of a specific set of high-quality plans rather than a single “optimal” one empowers clinicians with flexibility and supports clinical judgement by giving insights into interesting trade-offs for the patient at hand. At the same time, the customisable framework of BRIGHT makes it a valuable platform for research and innovation that may offer a vehicle by which to understand and unify practices across institutions.

 

Leah R.M. Dickhoff

Leiden University Medical Centre

l.r.m.dickhoff@lumc.nl