Prostate brachytherapy optimization using GPU accelerated simulated annealing and Monte Carlo dose simulation

Planning the radioactive seeds delivery during prostate brachytherapy is a critical part of the overall procedure. The planning process is time consuming and requires substantial user input and implication to ensure the optimal decision on the seeds' locations. Therefore efforts have been done to help in the decision making and minimize the overall planning required time, introducing automatic optimization techniques. The principal idea of these techniques is the construction and minimization of a cost function considering certain prescribed parameters. By minimizing the cost function, the optimal seeds distribution can be retrieved. Therefore a successful minimization algorithm has to be able to search randomly the given solution space and find the global minimum, escaping existing local minima. Pouliot J. et al. [1] have successfully adopt the simulated annealing (SA) technique in the treatment planning optimization ofprostate brachytherapy. This approach is able to obtain clinically acceptable seed distributions after 20000 iterations within 15 minutes, time duration which is acceptable for treatment planning purposes prior to operation. However, the dose calculation using standard protocols induces significant uncertainties and the optimization result is limited by the dose calculation accuracy. GGEMS-Brachy, a framework using GPU accelerated Monte Carlo (MC) methods has been proposed to address the limitations of current dosimetric protocols by Lemarechal Y. et al. [2]. Within this context one can produce a dose calculation of 2% uncertainty in 9.35s / 2.5s using one or four GPUs respectively. Our goal is to combine the MC dosimetry accuracy delivered by GGEMSBrachy with the optimization procedure of SA to improve the dosimetric outcome for intraoperative radiotherapy procedures. In addition, we propose a simple yet efficient modification in the SA algorithm [1] to further decrease the computational cost of the optimization process exploiting the GPU capabilities in order to facilitate the introduction of MC simulation in treatment planning optimization.