Evolutionary bi-objective optimization for high-dose-rate prostate brachytherapy With a focus on robust catheter positions

A possible treatment of prostate cancer is high-dose-rate brachytherapy, which is a form of internal radiation therapy. A number of very thin needles, called catheters, are placed inside the prostate. Each catheter has a fixed set of positions, where a radioactive source can stay for certain amounts of time, called dwell times. In this thesis, a novel method was developed for optimization of catheter positions with the corresponding dwell times for high-dose-rate prostate brachytherapy.
The goal of the optimization is to obtain both a high dose in the target and a low dose in the organs at risk. Due to the conflicting nature of these criteria, a bi-objective optimization model (with two objectives) was used. For solving computationally difficult bi-objective problems, evolutionary algorithms are the state-of-the-art, such as MO-RV-GOMEA. By applying MO-RV-GOMEA and parallelizing it on a GPU, high-quality treatment plans for a patient could be obtained in 5 minutes.
When optimizing treatment plans, there are a number of uncertainties involved. Two types of uncertainties were studied, related to the reconstruction of the organ shapes and the placement of the catheters. These uncertainties were taken into account in the optimization problem, i.e., robust optimization was performed. Resulting treatment plans were robust both to organ reconstruction settings and small catheter displacements.
The method was tested on ultrasound data obtained at the time of catheter placement. Compared to literature, an improvement of up to 5% in the dose-volume indices could be achieved.