ESTRO38 Congress Report

4. ESTRO-ELEKTA BRACHYTHERAPY AWARD

Bi-objective optimization of dosimetric indices for HDR prostate brachytherapy within 30 seconds Anton Bouter, Tanja Alderliesten, Bradley R. Pieters, Arjan Bel, Yury Niatsetski, Peter A.N. Bosman

Centrum Wiskunde & Informatica, The Netherlands

Context of the study Themain difficulty in prostate HDR brachytherapy is finding a treatment plan, such that the prostate (and potentially other target volumes) receive at least a certain amount of radiation dose, while surrounding healthy organs receive as little radiation dose as possible. For this reason, some formof automated treatment planning is generally used to optimize a treatment plan. When possible, such a treatment plan should adhere to a clinical protocol that is defined in terms of dosimetric indices, which describe the dose received by a specific organ. Conventional automated treatment planning methods do not directly optimize dosimetric indices, as they are relatively time consuming to accurately calculate. Overview of abstract We used a previously introduced bi-objective (see Figure 1) treatment planning method that directly optimizes dosimetric indices, and results in a large set of patient- specific treatment plans with different trade-offs between target coverage and organ sparing. This allows a clinician to consider many optimized treatment plans, and select the most appropriate plan for the given patient. Plans obtained by this optimization method were previously considered by physicians to be preferable tomanually optimized plans but required one hour of computation time. In this work, we aim to reduce the required computation time to the extent that it can be used in clinical practice.

What were the three main findings of your research? Mainly due to parallelization on a Graphics Processing Unit (GPU), our bi-objective treatment planning method now requires as little as 30 seconds of computation time. Optimizing for more than 30 seconds did not substantially improve results. Clinical plans obtained by experienced planners using clinically available software IPSA/HIPO, followed by graphical optimization, satisfied all clinical criteria for only 4 out of 18 patients. Our optimization found plans satisfying all clinical criteria for 15 cases, including these 4. Results of 10 optimization runs (to show variation) are shown in Figure 2.

Figure 2: The leftmost subfigure shows, for 10 optimization runs, the two optimization objectives: Least Coverage Index (LCI) and Least Sparing Index (LSI). These objectives indicate the dosimetric indices that have the largest deviation from the clinical protocol. The rightmost two figures show dosimetric indices for a given LCI and LSI. What impact could your research have? Our research could lead to better, more insightful brachytherapy treatment planning, as it allows for fast computation and insightful comparison of many high-quality treatment plans with different trade-offs between target coverage and organ sparing. Furthermore, this research could lead to class solutions becoming obsolete for HDR prostate brachytherapy planning, as well as potentially other types of treatments in the future, because our optimization method is patient-specific, yet requires no tuning of weights, as opposed to various clinically used automated treatment planning methods. Is this research indicative of a bigger trend in oncology? Along withmany advances in the fields of optimization and machine learning, an ever-increasing amount of available computing power leads to many new opportunities in the field of radiation oncology. The evolutionary optimization methods used in our work offer unique opportunities, but are generally deemed too time-consuming in practice. However, by employing state-of-the-art evolutionary algorithms and tailoring them to effectively exploit the capability of the latest high-performance (consumer-grade) hardware, these methods are now becoming time-efficient enough for clinical practice.

Figure 1: The result of our bi-objective treatment planning method is a set of non-dominated treatment plans. Each circle/cross indicates a treatment plan. Plans in the green area fulfil all sparing and coverage evaluation criteria of the clinical protocol.

AWARDS | Congress report

52