Evaluation of hybrid inverse planning and optimization (HIPO) algorithm for optimization in real-time, high-dose-rate (HDR) brachytherapy for prostate

Abstract

The purpose of this study is to investigate the effectiveness of the HIPO planning and optimization algorithm for real-time prostate HDR brachytherapy. This study consists of 20 patients who underwent ultrasound-based real-time HDR brachytherapy of the prostate using the treatment planning system called Oncentra Prostate (SWIFT version 3.0). The treatment plans for all patients were optimized using inverse dose-volume histogram-based optimization followed by graphical optimization (GRO) in real time. The GRO is manual manipulation of isodose lines slice by slice. The quality of the plan heavily depends on planner expertise and experience. The data for all patients were retrieved later, and treatment plans were created and optimized using HIPO algorithm with the same set of dose constraints, number of catheters, and set of contours as in the real-time optimization algorithm. The HIPO algorithm is a hybrid because it combines both stochastic and deterministic algorithms. The stochastic algorithm, called simulated annealing, searches the optimal catheter distributions for a given set of dose objectives. The deterministic algorithm, called dose-volume histogram-based optimization (DVHO), optimizes three-dimensional dose distribution quickly by moving straight downhill once it is in the advantageous region of the search space given by the stochastic algorithm. The PTV receiving 100% of the prescription dose (V100) was 97.56% and 95.38% with GRO and HIPO, respectively. The mean dose (D(mean)) and minimum dose to 10% volume (D10) for the urethra, rectum, and bladder were all statistically lower with HIPO compared to GRO using the student pair t-test at 5% significance level. HIPO can provide treatment plans with comparable target coverage to that of GRO with a reduction in dose to the critical structures.

Figure 1

HIPO starts with a simulated annealing algorithm to find the optimal catheter distribution for a given anatomy and user defined number of catheters, objectives, dosimetric constraints and penalties. As trials progress, HIPO (a) places a catheter in a feasible but unoccupied template hole randomly and the superiority of the resulting catheter distribution is tested based on the aggregate objective function (b) normalized to its initial (first trial) value stochastically. If the user does not interrupt, the algorithm runs for the predefined number of trials.

Figure 2

Once HIPO finds an optimal catheter distribution (a) using the simulated annealing algorithm, it then takes this as an initial input. Then the DVHO algorithm optimizes the 3D dose distribution for the given catheter distribution, objectives, dosimetric constraints, and penalties deterministically. DVHO algorithm (b) is guided to generate an ideal DVH and tries to reduce hot areas (red), or cold areas (blue) in the case of the PTV, and tries to reduce the upper dose limit for each OARs deterministically.

Figure 3

A comparison of the isodose distribution in colorwash form between GRO (a) and HIPO (b) for the same axial slice. The PTV is enclosed with a red contour, $\text{blue}=100\%$ isodose, $\text{green}=125\%,\text{brown}=150\%,\text{light coral}=200\%$. Normal tissue treated outside the PTV is less with HIPO than with GRO.

Figure 4

A DVH comparison between GRO (dotted lines) and HIPO (solid lines) for a typical case. HIPO yields smaller dose to the bladder and rectum compared to GRO with comparable dose to urethra and prostate.

Figure 5

Box and whisker plots of the $\text{PTV}{\mathrm{D}}_{90}\left(\mathrm{a}\right)$ and ${\mathrm{V}}_{100}$ (b) for the four optimization algorithms covered in this study. DVHO has the widest range of ${\mathrm{D}}_{90}$ and ${\mathrm{V}}_{100}$ rendering clinically unacceptable most of the time. The plots present the 10th percentile, 25th percentile, median, 75th percentile, and 90th percentile of data used.

Figure 6

Box and whisker plots of the $\text{PTV}{\mathrm{V}}_{150}\left(\mathrm{a}\right)$ and ${\mathrm{V}}_{200}$ (b) for GRO, HIPO1, and HIPO2. Both are significantly larger with GRO.

Figure 7

Box and whisker plots of COIN (a) and HI (b) for GRO, HIPO1, and HIPO2. Both are bigger with HIPO1 as compared to GRO. HIPO2 has comparable COIN and HI to that of HIPO1.

Figure 8

Box and whisker plots of ${\mathrm{D}}_{10}$ for critical structures urethra (a), bladder (b), and rectum (c) for GRO, HIPO1, and HIPO2. All are greater with GRO compared to HIPO1. There is no significant change in ${\mathrm{D}}_{10}$ for each structure between HIPO2 and HIPO1.

Figure 9

Box and whisker plots of ${\mathrm{D}}_{\text{mean}}$ for the urethra (a), bladder (b), and rectum (c) for GRO, HIPO1, and HIPO2. Mean doses to all critical structures are larger with GRO compared to HIPO1 and are comparable between HIPO1 and HIPO2.