GPU‐accelerated bi‐objective treatment planning for prostate high‐dose‐rate brachytherapy

Abstract

Purpose The purpose of this study is to improve upon a recently introduced bi‐objective treatment planning method for prostate high‐dose‐rate (HDR) brachytherapy (BT), both in terms of resulting plan quality and runtime requirements, to the extent that its execution time is clinically acceptable. Methods Bi‐objective treatment planning is done using a state‐of‐the‐art multiobjective evolutionary algorithm, which produces a large number of potential treatment plans with different trade‐offs between coverage of the target volumes and sparing organs at risk. A graphics processing unit (GPU) is used for large‐scale parallelization of dose calculations and the calculation of the dose‐volume (DV) indices of potential treatment plans. Moreover, the objectives of the previously used bi‐objective optimization model are modified to produce better results. Results We applied the GPU‐accelerated bi‐objective treatment planning method to a set of 18 patients, resulting in a set containing a few hundred potential treatment plans with different trade‐offs for each of these patients. Due to accelerations introduced in this article, results previously achieved after 1 hour are now achieved within 30 seconds of optimization. We found plans satisfying the clinical protocol for 15 of 18 patients, whereas this was the case for only 4 of 18 clinical plans. Higher quality treatment plans are obtained when the accuracy of DV index calculation is increased using more dose calculation points, requiring still no more than 3 minutes of optimization for 100 000 points. Conclusions Large sets of high‐quality treatment plans that trade‐off coverage and sparing are now achievable within 30 seconds, due to the GPU‐acceleration of a previously introduced bi‐objective treatment planning method for prostate HDR brachytherapy. Higher quality plans can be achieved when optimizing for 3 minutes, which we still consider to be clinically acceptable. This allows for more insightful treatment plan selection in a clinical setting.

Fast and insightful bi-objective optimization for prostate cancer treatment planning with high-dose-rate brachytherapy

“…While our approach can be considered sufficiently fast for dwell-time optimization tasks on computer desktops with only CPU cores in current clinical practice, real-time HDR-BT planning problems in the upcoming future, e.g., MR-guided catheter placement [37], would require a faster response time. A potential solution is to further accelerate MO-RV-GOMEA with the use of GPU parallelization, which we recently successfully achieved [38]. To further speed up both the CPU and GPU versions of our approach, we will study alternative ways of sampling dose calculation points that, for the specific DV indices of interest, result in the same precision of the DV indices, but using fewer dose calculation points (e.g., by exploiting shapes and surfaces of organs) [39,40].…”

Bi‐objective optimization of catheter positions for high‐dose‐rate prostate brachytherapy

“…The GPU parallelization of the simultaneous optimization of catheter positions and dwell times using GOMEA expands previous work focused only on dwell time optimization. 5 In this approach, sets of treatment plans are evaluated in parallel, and the dose in each dose-calculation point of each of these treatment plans is calculated on a separate thread. The programming for the GPU was done in CUDA (NVIDIA Corporation, Toolkit v8.0.61) and was run on an NVIDIA Titan Xp GPU, which contains 12 GB of memory.…”

“…Hence, to achieve the shortest runtime of the GPU parallelization, all subsets with <5 variables are removed from the FOS, following previous work. 5 For catheter position optimization, we compute the distance between catheters and between dwell positions by first computing the average catheter positions over the population. For the two different types of variables, namely catheter position variables and dwell time variables, these distances are used to build two separate Linkage Trees.…”

“…These are the numbers of dose-calculation points with which the bi-objective dwell time optimization approach was introduced and validated, 12,25 and for which the difference between optimized and re-evaluated values was shown to be small. 5 For healthy tissue, optimization was already performed on 20.000 dose-calculation points, to avoid slight violations of the constraint during re-evaluation. The healthy tissue was defined as the smallest parallelepiped containing all active dwell positions in the catheter grid, with an added 5 mm margin, excluding all delineated organs (and tumor).…”

“…For this, current practice is to optimize the dwell times using automated methods such as the inverse planning simulated annealing (IPSA) 2 or more recently the hybrid inverse planning optimization (HIPO) 1 algorithms. Other methods have been investigated for optimizing the dwell times directly on dose-volume indices, [3][4][5] as well as (dose-volume based) approaches that consider multiple objectives. [6][7][8][9][10] A multi-objective optimization method that directly optimizes dose-volume indices has been proposed by leveraging the Multi-Objective Real-Valued Gene-pool Optimal Mixing Evolutionary Algorithm (MO-RV-GOMEA).…”