Assessment of specific versus combined purpose knowledge based models in prostate radiotherapy

Abstract Knowledge‐based planning (KBP) can be used to improve plan quality, planning speed, and reduce the inter‐patient plan variability. KPB may also identify and reduce systematic variations in VMAT plans, something very important in multi‐institutional clinical trials. Training of a KBP library is a complex and difficult process, and models must be validated prior to their clinical use. The purpose of this work is to assess the quality of the treatment plans generated using a specific versus combined purpose model KBP library for prostate cancer. Seven KBP model libraries were created from a set of patients treated on various Institutional Review Board (IRB) approved protocols. All KBP libraries were validated using an independent set of twenty patients (half treated Pr: Prostate alone half treated PLN: prostate plus pelvic lymph nodes). Two models were tested on the Pr patients only, four tested on PLN patients only, and one tested on all patients. All plans were normalized such that at least 95% of the prostate planning target volume received 100% of the planned dose. The plans based on different model libraries were compared to each other and the expert clinical plan. For Pr plans there were almost no statistically significant differences (P < 0.008) between the plans types except conformity index (CI) with library plans better than the expert. For PLN plans, all model libraries in generally showed femur doses and CI better than the expert plans (P < 0.003). This study demonstrated that no large differences were observed between specific versus combined KBP model libraries in dosimetry of prostate cancer patients. This would allow for a fewer specific plans to be needed to create a model library. Further studies are needed to evaluate benefits of combined purpose model libraries for planning of complex sites such as head and neck cancer.


| INTRODUCTION
Both intensity modulated radiotherapy (IMRT) and volumetric modulated arc therapy (VMAT) are advanced technologies that have been commonly used for treatment of prostate cancer. 1,2 However, there may be large variations in the quality of the treatment plans due to differences in both experience and skill of the treatment planners.
Such differences may limit the desired organs-at-risk (OAR) sparing and the target coverage that can be achieved. Recent investigations have demonstrated that knowledge based planning (KBP), which utilizes a library of previously treated patient plans, has enormous potential for improving the quality and consistency of treatment plans. [1][2][3][4][5][6][7][8] KBP library models are a way to objectively incorporate years of prior solid expert planning experience into the planning process. KBP allows planners of all experience levels to predict the best possible planning goals as well as to create treatment plans that draw on the lessons from successful prior plans. 3,4 If created properly, such libraries have the power and potential to shorten the time it takes to produce high quality treatment plans as well as to making it possible for planners at all experience levels to generate such plans. A tremendous amount of work goes in to creating these models. Many things must be considered in choosing plans for the models including the quality of the contours, the variations in size for the organs at risk (OARs) and target volume(s), the types of treatment plan (including field setup, energy used, and treatment technique), the overall quality of the plan, the dose goals reached by the plan, and finally the prescription level or levels used in creation of the plan. 9 Initial evaluation of KBP has demonstrated that KBP is able to generate clinically acceptable plans for the treatment of prostate cancer. 1,10 However, a question arose during the creation of models for prostate planning as to the appropriateness of including plans of more than one type in the model. For instance, does creating a model containing plans that treat prostate alone alongside plans that treat prostate plus pelvic lymph nodes outperform a model that includes only one of those two plan types? A search of the literature showed no answer as to whether it was either preferable or necessary to use plans of a single type in the model in order to get the best results. The drawback of only being able to use a single plan type is clear: one would need to have a large number of single purpose plans in order to create a viable model. Being able to include plans of more than one type means that an initial model can be created with fewer plans overall. This study seeks to answer the question of whether the use of a combination of different plan types in the creation of KBP model libraries produces plans of the same quality when compared to a KBP model that consists only of plans of a specific type.

2.A | Patient selection
Ninety-seven Volumetric Modulated Arc Therapy (VMAT) plans were selected from a database of patients who were enrolled in various IRB-approved protocols for treatment of prostate alone (Pr) or prostate plus pelvic lymph nodes (PLN). For all prostate patients on these protocols, RapidArc (ver. 13

2.C | Model validation
Twenty patients (10 Pr alone, and 10 PLN) that were not included in any of the KBP library-training sets were used for model validation.
Models (a) and (d) were validated using the ten Pr alone patients, models (b), (e), (f) and (g) were validated using ten patients treated to the prostate plus lymph nodes, and model (c) was validated using all 20 patients. Validation means that a new treatment plan was generated for each of these patients utilizing a single run of the RapidPlan optimizer with minimal planner intervention utilizing each appropriate KBP library. All treatment plans were examined for quality with both physicians and physicists comparing qualitatively and quantitatively the plans.

2.D | Plan evaluation
The plans generated using different KBP libraries were compared to each other and the clinical plans using the PTV dose coverage and OAR sparing based on the dose-volume parameters listed in Table 1.
Specifically, the data points taken on each plan were the minimum, where PIV is the prescription isodose volume and TV is the tumor volume. 13 The HI used here is defined as where D 2% is the dose to 2% of the PTV, D 98% is the dose to 98% of the PTV and D p is the prescription dose for the PTV. 14

| RESULTS
All plans generated by the original KBP model libraries were considered clinically acceptable for treatment within the guidelines set out in the treatment protocols, meaning they met or exceeded the goals in Table 1. Tables 2 and 3 list the plan comparisons for Pr alone and PLN, respectively, which showed statistical significance (P < 0.008 for prostate alone and P < 0.003 for prostate plus lymph nodes).
Note that on the list in Table 2  There were no statistically significant differences between in plan quality when comparing the large sized model (n = 97) and the smaller sized models (n = 66 or n = 31); as well as no difference between the models with different ratios of cases types for the various PPLN models.    indicates that the more focused model would not necessarily do better or worse, though the details of this question is left to future work.

| DISCUSSION
The information seen here shows that the broader model will do as well, and the number of plans mimicking a particular geometry would not need to be as high in order to create a robust KBP library capable of planning on a large body of cases.
Interestingly, the size of the models (66 vs 97 plan models, and 33 vs 97 plan models) made no significant difference in the plan quality for either the Pr or the PLN cases. This could be a case of quality inquality out. That is to say, the quality of the cases that made up the database were consistent and good enough as to render having 2 or 3 times the number of plans unnecessary. As well, for PLN cases the ratio of plans in the model made no difference in the model's ability to create a viable plan. Again, this possibly speaks to the quality of the plans in the model that a random sampling gave us good models in all cases.

| CONCLUSIONS
This study indicates that a combined KBP library model library performs as well as a single purpose model, especially for the more complex plans. This indicates that a good prostate cancer model can be created with a mix of plans for treating prostate alone and prostate plus pelvic lymph nodes, and this model will perform well, even for more complex treatment geometries. The general feeling is that this result could be extended to other body sites and plan types, though further investigation is warranted.

This work is funded by Varian Research Grant titled "Knowledge
Based Planning of Prostate Cancer."

CONF LICTS OF INTEREST
Dr. Elizabeth Bossart received travel support to attend the Varian Research Partners meeting from Varian Medical Systems.