Photon Optimizer (PO) prevails over Progressive Resolution Optimizer (PRO) for VMAT planning with or without knowledge‐based solution

Abstract The enhanced dosimetric performance of knowledge‐based volumetric modulated arc therapy (VMAT) planning might be jointly contributed by the patient‐specific optimization objectives, as estimated by the RapidPlan model, and by the potentially improved Photon Optimizer (PO) algorithm than the previous Progressive Resolution Optimizer (PRO) engine. As PO is mandatory for RapidPlan estimation but optional for conventional manual planning, appreciating the two optimizers may provide practical guidelines for the algorithm selection because knowledge‐based planning may not replace the current method completely in a short run. Using a previously validated dose–volume histogram (DVH) estimation model which can produce clinically acceptable plans automatically for rectal cancer patients without interactive manual adjustment, this study reoptimized 30 historically approved plans (referred as clinical plans that were created manually with PRO) with RapidPlan solution (PO plans). Then the PRO algorithm was utilized to optimize the plans again using the same dose–volume constraints as PO plans, where the line objectives were converted as a series of point objectives automatically (PRO plans). On the basis of comparable target dose coverage, the combined applications of new objectives and PO algorithm have significantly reduced the organs‐at‐risk (OAR) exposure by 23.49–32.72% than the clinical plans. These discrepancies have been largely preserved after substituting PRO for PO, indicating the dosimetric improvements were mostly attributable to the refined objectives. Therefore, Eclipse users of earlier versions may instantly benefit from adopting the model‐generated objectives from other RapidPlan‐equipped centers, even with PRO algorithm. However, the additional contribution made by the PO relative to PRO accounted for 1.54–3.74%, suggesting PO should be selected with priority whenever available, with or without RapidPlan solution as a purchasable package. Significantly increased monitor units were associated with the model‐generated objectives but independent from the optimizers, indicating higher modulation in these plans. As a summary, PO prevails over PRO algorithm for VMAT planning with or without knowledge‐based technique.


| INTRODUCTION
Using the anatomical structures, field geometry, dose metrics, and dose prescription of previous plans as historical experiences to predict the dose-volume objectives for the upcoming patients, 1,2 knowledge-based treatment planning has been deemed as a promising solution to reduce the subjective inter-planner varieties, 3-10 enhance the clinical efficiency and quality of prospective plans. 2,[11][12][13][14][15] As a commercial knowledge-based optimizer, RapidPlan in Eclipse treatment planning system (V13.5 or later, Varian Medical Systems, Palo Alto, CA, USA) has been validated by many publications suggesting superior dosimetric outcomes than the conventional methods. [16][17][18][19][20][21][22][23] However, in addition to the usage of possibly advanced personalized optimization objectives as generated by the dose-volume histogram (DVH) estimation models, the new Photon Optimizer (PO) was introduced to combine and substitute for the two old algorithms, that is, Progressive Resolution Optimizer (PRO) for volumetric modulated arc therapy (VMAT) and Dose Volume Optimizer (DVO) for static field intensity-modulated radiotherapy (IMRT). By far, no study has been conducted to confirm if the observed dosimetric progress was also partially (if any) attributable to the potentially better PO algorithm.
According to the manufacturer, DVO optimizes the field shape and intensity using a simple gradient optimization to approach the desired dose-volume objectives. The fluences are back-projected from the derivates of the costs at each cloud point representing the patient volume. PRO optimizes MLC leaf positions and MU/deg based on control points segmentation of the gantry angle. As the optimization progresses, the accuracy of the angle resolution and dose calculation segments increase. Relative to prior optimizers, PO has involved critical changes: the old point cloud model for PRO and DVO has been replaced by a new volume representation, where "structures, DVH calculation and dose sampling are defined spatially by using one single matrix over the image". However, DVO and PRO have a more powerful dose-volume objective form than the PO in terms of limiting the local minima in the optimization space. More technical details were described by Vanetti et al. 24 and Cozzi et al. 25 It is noticed that PO is mandatory for RapidPlan to accommodate its geometry-based expected dose (GED) algorithm, yet is optional for the conventional manual planning where all the optimization features can be used in the same way without a model.
Providing knowledge-based planning is not likely to replace the conventional methods completely in a short run, appreciating the behavior of the new PO against the old optimizers may provide useful guidelines for algorithm selection when RapidPlan is not invoked as a separate executable option. After all, as a purchasable engine, RapidPlan is not available to every Eclipse user, and the model configuration can be a sorely time-consuming process. In order to assess the different algorithms based on the same (but patient specific) optimization objectives that can produce clinically acceptable plans without interactive manual adjustment, all DVH constraints were derived from our RapidPlan model estimation that has been validated for rectal cancer patients. 16,17 However, this model was configured with features that are not supported by DVO, such as mean dose objective and automatic normal tissue objective; hence, this study only focuses on VMAT planning using PO and PRO algorithms, respectively.

| METHODS
This study was conducted based on Varian Eclipse Treatment Planning System V. 13.5.

2.A | DVH estimation model
As a brief summary of the previous work, 16,17 a DVH estimation model was configured using 81 historical VMAT plans for preoperative rectal cancer patients with simultaneous-integrated-boosting (SIB). The library size was determined based on Boutilier's study and the rule-of-thumb approach. 26 The gross target volume (GTV) was delineated to cover the primary tumor, mesorectal space, and the involved lymph nodes. 27 The clinical target volume (CTV) included GTV, presacral region, mesorectal/lateral lymph nodes, internal iliac lymph node chain, and pelvic wall area. Should the anterior organ involvement was suspected, CTV also includes the external iliac lymph nodes, and includes the inguinal lymph nodes when the lower third of the vagina was invaded or major tumor extension into the internal and external anal sphincter was observed. 28 Isotropic margins of 5 mm were applied to create planning target volumes (PTV boost from GTV, and PTV from CTV, respectively). Target dose of 50.6 Gy and 41.8 Gy were prescribed to 95% of PTV boost and PTV in 22 fractions, respectively. All plans were created with 1-2 full arc, 5°collimator rotation, and 10 MV photon beams modulated by Varian Millennium 120 multi-leaf collimator mounted on a Varian Trilogy accelerator. As reported before, the model validation on over 100 historical plans displayed significantly improved dosimetric results than the average clinical level, after applying the RapidPlangenerated objectives and the PO algorithm.

2.B | VMAT planning using PO and PRO
Thirty clinical VMAT rectal plans that have been manually optimized with PRO were retrospectively selected and reoptimized using the RapidPlan engine (referred as PO plans). Representing the average planning quality, these 30 consecutively treated plans were not used for model development but followed consistent contouring 28  . This approach ensured the clinically acceptable PO, and PRO plans were derived from identical constraints and objectives without interactive human adjustment during the optimization process. All other parameters were maintained as earlier during the optimization as well; hence, the observable discrepancies were mostly ascribed to the disparities of PO and PRO algorithms. Renormalization was performed to assess the organs-at-risk (OAR) sparing based on comparable target dose coverage (i.e., 95% PTV boost and PTV were covered by corresponding prescribed dose, respectively).

2.C | Dosimetric evaluation and statistical method
The clinical, PO and PRO plans were evaluated mutually by means of the target homogeneity index (HI PTVboost and HI PTV ); the target conformity index (CI PTVboost and CI PTV ); dose to 50% of the volume for the femoral head and urinary bladder (D 50%_FH and D 50_UB ), the mean dose (D mean_FH and D mean_UB ); the hot spot volume receiving over 107% of the prescribed dose to PTV boost (V 107% , i.e., V 54.14Gy ), and the total monitor units (MU). Based on SPSS 21 software (IBM Analytics, Armonk, NY, USA), paired samples t-test was used to analyze the normally distributed data (Shapiro-Wilk method); otherwise Wilcoxon signed-rank test was conducted. Significant level was put at P < 0.05 (2-tailed).

| RESULTS
Without interactive human interference, the patient-specific optimization objectives generated from the model estimation functioned well with both PO and PRO algorithms, which produced clinically acceptable plans automatically as visually inspected on three-dimensional dose color wash and DVH distribution.  Taking the clinical plans as baseline, Table 3 breaks down the relative improvement of OAR sparing as contributed by the new objectives and the optimizers, which differs by an order of magnitude.
T A B L E 1 Dosimetric comparison (target coverage) of 30 patients that were planned by: PRO using the manual objectives as in the clinical plans (clinical); PO using the RapidPlan-generated objectives (PO); and PRO using the RapidPlan-generated objectives (PRO). Dose unit (Gy).  T A B L E 2 Dosimetric comparison (OAR sparing) of 30 patients that were planned by: PRO using the manual objectives as in the clinical plans (clinical); PO using the RapidPlan-generated objectives (PO); and PRO using the RapidPlan-generated objectives (PRO). Dose unit (Gy).   Computation-based analysis has excluded the inferior deliverability of knowledge-based plans, 14 yet physical measurement should still be desirable to gain more confidence. In addition, when more DVH estimation models become available, the comparative assessment could be performed for DVO algorithm and for other lesion sites.

Mean
This study is potentially limited by the nondeterministic nature of the optimization process: slightly different results may be obtained even if the same algorithm and constraints were utilized.
However, this random and minor uncertainty should be largely canceled out by the averaging during the statistical analysis, and the systematic discrepancies should be mostly attributable to the different optimizers.