Evaluation of auto‐planning in IMRT and VMAT for head and neck cancer

Abstract Purpose The purposes of this work are to (a) investigate whether the use of auto‐planning and multiple iterations improves quality of head and neck (HN) radiotherapy plans; (b) determine whether delivery methods such as step‐and‐shoot (SS) and volumetric modulated arc therapy (VMAT) impact plan quality; (c) report on the observations of plan quality predictions of a commercial feasibility tool. Materials and methods Twenty HN cases were retrospectively selected from our clinical database for this study. The first ten plans were used to test setting up planning goals and other optimization parameters in the auto‐planning module. Subsequently, the other ten plans were replanned with auto‐planning using step‐and‐shoot (AP‐SS) and VMAT (AP‐VMAT) delivery methods. Dosimetric endpoints were compared between the clinical plans and the corresponding AP‐SS and AP‐VMAT plans. Finally, predicted dosimetric endpoints from a commercial program were assessed. Results All AP‐SS and AP‐VMAT plans met the clinical dose constraints. With auto‐planning, the dose coverage of the low dose planning target volume (PTV) was improved while the dose coverage of the high dose PTV was maintained. Compared to the clinical plans, the doses to critical organs, such as the brainstem, parotid, larynx, esophagus, and oral cavity were significantly reduced in the AP‐VMAT (P < 0.05); the AP‐SS plans had similar homogeneity indices (HI) and conformality indices (CI) and the AP‐VMAT plans had comparable HI and improved CI. Good agreement in dosimetric endpoints between predictions and AP‐VMAT plans were observed in five of seven critical organs. Conclusion With improved planning quality and efficiency, auto‐planning module is an effective tool to enable planners to generate HN IMRT plans that are meeting institution specific planning protocols. DVH prediction is feasible in improving workflow and plan quality.


| INTRODUCTION
Head and neck (HN) cancer is a technically challenging treatment site in radiation oncology due to the complex anatomy and numerous organs at risk (OARs) in close proximity to targets. Treatment planning techniques for HN cancer have advanced from the conventional three-field technique to intensity modulated radiation therapy (IMRT) over two decades. 1 To achieve adequate target coverage while protecting numerous OARs, IMRT plans for HN cancer require highly conformal dose distributions and a steep dose fall-off between the boundary of tumor volumes and sensitive structures. With limited clinical resources (time and manpower), a major challenge in HN IMRT planning is large variations in plan quality among treatment planners in part due to varied planning skills and limited planning time. [2][3][4] Many publications have identified variations in IMRT plan quality. Hunt et al. 5  Allowing flexibility and patient-specific organ sparing prediction, a commercial product, PlanIQ Feasibility (Sun Nuclear Corp., Melbourne, FL), has been developed. The predicted dose volume histograms (DVHs) are based on energy-specific dose spread calculation, reflecting the characteristics of photon dose distribution in media. 8 Another approach to robust planning is to create many planning solutions (multicriteria optimization) for a single clinical case so that clinicians can make a decision based on the trade-off among the dose coverage of the tumor volume and protections of sensitive structures. 9 The automatic planning tool developed by the Pinnacle (Philips Radiation Oncology Systems, Fitchburg, WI) commercial treatment planning system is to mimic the manual processes of skilled planners by progressively and iteratively adjusting and adding planning objectives, which may mitigate the shortcoming of the gradient-based optimization. 10 In an ideal world, a planner would be equipped with all of these tools: a tool that can reliably predict achievable DVHs as initial inputs of the planning objectives, a tool that can automatically and progressively adjust planning objectives, and a tool that can offer multiple solutions based on different trade-offs.
The purposes of this study are to (a) investigate whether the use of automation and multiple iterations can improve quality of HN plans; (b) determine whether delivery methods such as step-andshoot (SS) and volumetric modulated arc therapy (VMAT) impact plan quality; (c) report on the observations of auto-plan qualities with respect to the prediction of the feasibility tool.

2.A | Patient selection
Twenty HN patients with various tumor sites and stages were retrospectively selected from an institutional review board approved registry. We purposely chose these patients to reflect various clinical scenarios. The first ten of the twenty patient plans were used for initial testing of the auto-planning model in Pinnacle. These patients were treated with nine-beam step-and-shoot IMRT plans for either definitive or postoperative intent. For definitive cases, the primary targets were prescribed to a dose of 70-72 Gy while the regional lymph nodes were prescribed to a dose of 54-58 Gy. For post-operative cases, the prescription doses to the primary tumor beds were 60-66 Gy. The details of the tumor locations, stages, and prescription doses for the second set of ten HN patients are listed in Table 1.

2.B | HN planning goals
The general HN planning goals and plan acceptance criteria have been established in our department. The treatment goals were to deliver prescription doses to ≥95% of the high dose planning target volumes (HD_PTV) and ≥95% of the low dose planning target volumes (LD_PTV). The planning acceptance criteria for OARs are listed in Table 2. Planners adjusted the planning goals for individual cases in consultation with the attending physicians due to variability in the anatomic relationship between PTVs and OARs across each case.

2.C | Auto-planning module
A commercial auto-planning module from the Pinnacle 3 treatment planning system (Pinnacle 3 9.10, Philips Healthcare Inc., Fitchburg, WI) was clinically implemented in our institution in January 2015.
Prior to clinical implementation, we validated this tool by comparing the second set of ten clinical HN plans to step-and-shoot IMRT plans and two-arc volumetric modulated arc therapy (VMAT) plans generated from the auto-planning module on the same patient data sets.
All clinical plans used nine equally spaced beams, 6 MV photon energy, and step-and-shoot delivery method. The step-and-shoot IMRT auto-plans (AP-SS) used the same beam angles as the clinical plans. Since our practice has transitioned to VMAT delivery for most HN patients, we also compared two-arc VMAT auto-plans (AP-VMAT) of these patients with their clinical plans.
In manual planning for HN cancer, typically more than 40 planning objectives and their associated numerical weights are entered by the planner. With the auto-planning module, simplified planning goals as shown in Table 3 are entered by the planner. Based on the user input in the planning goals, the auto-planning module then creates detailed planning objectives. The auto-planning module uses an iterative process to mimic the manual planning process by separating overlapped contours, creating tuning structures, adjusting hot and cold spots, and optimizing conformality and homogeneity. After auto-planning is completed, planners can further manually adjust the planning objectives and continue the "warm start" optimization as they often do during manual optimization. Or the planners can reset all beams, adjust planning goals, and start the auto-planning process from the beginning. To further automate the treatment planning, users may create a site-specific planning technique or a class solution, saved as a technique into the institution's library, which defines common planning parameters such as prescriptions, beam angles, beam energy, and treatment machine.
For the purpose of this study, we created two HN specific techniques: one used the nine beam step-and-shoot delivery, and the other used two VMAT arc delivery. Both techniques used the same planning goals for normal structures. Since our institution uses multiple machines to treat HN patients and some treatment machines do not have VMAT delivery, planners still must choose a specific treatment machine after loading the HN specific planning technique.
For the nine-beam AP-SS, the direct machine parameter optimization (DMPO) was chosen and the two-arc AP-VMAT, the optimization type chosen was the "SmartArc" from the Pinnacle system with the dose calculation at every 4˚with a convolution and superposition algorithm. For each HN case selected for this study, three plans were created: one clinical plan, one AP-SS, and one AP-VMAT.

2.D | Plan evaluation
Plan quality was evaluated based on several dosimetric endpoints for PTVs and critical structuresincluding dose volume coverage, maximum dose to 0.03 cc (D 0.03cc ), and mean dose (D mean )as well as the conformality index (CI), the homogeneity index (HI), and the total monitor units (MUs) per fraction. The CI 11 was defined as where V Rx is the tissue volume covered by the prescription dose for the HD_PTV and V PTV is the volume of the HD_PTV. For the ideal case, CI = 1. The HI was defined as where D max is the maximum dose of the plan and D Rx is the prescription dose for the HD_PTV. The total MUs per fraction were also used to assess the plan delivery efficiency.
T A B L E 2 Head and neck planning acceptance criteria. factor is defined as the feasibility factor, with higher feasibility associated with higher f. The estimation is based on a series of energyspecific dose spread calculations, independent of any particular beam arrangement. 8 For a specific patient, this estimated calculation is based on the heterogeneous dataset along with the geometric relationship between the targets and OARs while taking into account the high-(penumbra driven) and low (PDD and scatter-driven) gradient dose spreading. The predicted DVHs from PlanIQ can be used as the input of IMRT planning objectives or as a tool for quality assurance. In this paper, we use the predicted DVHs for the latter.

2.F | Statistical analysis
One sided paired sign test was used to test the difference in medians of the dosimetric endpoints between the clinical plans and the corresponding AP-SS and AP-VMAT plans. 12 The test is conducted by subtracting the paired values from two groups and counting the positive (c+) or negative (c−) signs. Let c equal the smaller one of c+ and c−, and let N be the total number of unequal pairs. The P-value is given by the cumulative binomial distribution, The one sided test was used under the null hypothesisthe AP-SS/AP-VMAT plans are not better than the clinical plans in compared items. Statistical significance is achieved when P < 0.05 to conclude that the AP-SS/AP-VMAT plans are better than the clinical plans.
The Spearman rank correlation coefficient is used to describe the monotonic association between the PlanIQ feasibility and AP-VMAT endpoints. 13 The correlation coefficient is given by the following equation, where Δd i is the difference between the ranks for each pair, and N is the total number of pairs.
The correlation coefficients were interpreted as: very high correlation if r s > 0.9; high correlation if 0.7 < r s ≤ 0.9; moderate correlation if 0.5 < r s ≤ 0.7; low correlation if 0.3 < r s ≤ 0.5; negligible correlation if r s ≤ 0.3. Figure 1 shows the median, interquartile range (IQR), minimum, and maximum values of the selected dosimetric endpoints, HI, CI, and MU for the second set of ten HN patients. All AP-SS plans and AP-VMAT plans met the clinical dose limit requirements. As shown in For a selected patient, Fig. 2 show that for lung stereotactic body radiotherapy auto-planning reduces optimization time by 77.3% and total monetary cost by 3.6%. 14 As reported by Creemers et al., auto-planning requires roughly the same total planning time compared to manual planning, but it reduces the planners' "hands-on-time" by 75%. 15 The auto-planning module has some limitations. The beam arrangement must be initially set and cannot be changed during auto-planning. Auto-planning runs six optimization iterations, which may be excessive for simple cases. Though auto-planning techniques used in this study generated clinically acceptable plans for all ten HN patients without further modification, other patient cases may still require manual adjustments to achieve optimal results.

| RESULTS
In this work, nonparametric statistical tests such as sign test and Spearman rank correlation are used due to the small sample size.
One must cautiously interpret the results as they are not as powerful as parametric tests such as t-test and Pearson correlation. We study plans with two prescription dose levels to maintain the data homogeneity. However, three dose level HN plans are also common at other institutions while our institution has adopted to two dose levels for most patients with HN cancer. With more prescription levels, the geometric and dosimetric relationships between targets and OARs will change, which may affect the auto-planning. Adjustment in the auto-planning technique is needed to accommodate such prescription changes even for the same disease site.
Auto-planning, among other methods such as knowledge-based planning and multicriteria optimization, is one of the advanced planning techniques to improve planning consistency and efficiency. HN cancer is one of the most challenging sites for treatment planning, and Pinnacle auto-planning is confirmed as a viable solution in an early study. 16 Other sites, such as prostate, 17 esophagus, 18 lung, 14,15 and brain, 19 are also investigated by different groups. All studies have confirmed that auto-planning generates clinically acceptable T A B L E 4 Plan quality endpoints of the Clinical, AP_SS and AP_VMAT plans. One sided paired sign tests were performed between the Clinical and AP_SS, and between the Clinical and AP_VMAT. Results with P < 0.05 indicated statistical significance and were labeled with "*".  to the parotids has a range, the AP-VMAT plans mostly cluster around 25 Gy, which is reflective to the clinical requirements (Table 2) and auto-planning technique (Table 3). Although the AP-VMAT plans do not meet the predicted spinal cord dose [ Figure 3(b)], they all meet the clinical requirements, and they are compromised so that other OARs like the parotids also meet the requirements. The feasibility prediction assumes isotropic dose fall off rate surrounding the target, while in reality dose is designed to fall off differentially based on the importance and difficulty of the OAR constraints in each direction.

| CONCLUSION S
Nine-beam step-and-shoot and two-arc VMAT treatment planning techniques are developed using Pinnacle auto-planning for HN conventional radiotherapy. This auto-planning tool is promising in reducing clinical workload and improving plan quality. DVH predictions with PlanIQ feasibility show good agreement with AP-VMAT plans in the initial testing. Further study is warranted in order to fully implement the prediction tool for clinical use.

ACKNOWLEDG MENTS
This work was presented in part at the 57th Annual Meeting of the American Society for Radiation Oncology (ASTRO), San Antonio, TX,

Karl
Bzdusek is an employee of Philips Healthcare Inc. Ping Xia received a research grant from Philips Healthcare Inc. Others: None.