Automatic treatment planning for VMAT‐based total body irradiation using Eclipse scripting

Abstract The purpose of this work is to establish an automated approach for a multiple isocenter volumetric arc therapy (VMAT)‐based TBI treatment planning approach. Five anonymized full‐body CT imaging sets were used. A script was developed to automate and standardize the treatment planning process using the Varian Eclipse v15.6 Scripting API. The script generates two treatment plans: a head‐first VMAT‐based plan for upper body coverage using four isocenters and a total of eight full arcs; and a feet‐first AP/PA plan with three isocenters that covers the lower extremities of the patient. PTV was the entire body cropped 5 mm from the patient surface and extended 3 mm into the lungs and kidneys. Two plans were generated for each case: one to a total dose of 1200 cGy in 8 fractions and a second one to a total dose of 1320 cGy in 8 fractions. Plans were calculated using the AAA algorithm and 6 MV photon energy. One plan was created and delivered to an anthropomorphic phantom containing 12 OSLDs for in‐vivo dose verification. For the plans prescribed to 1200 cGy total dose the following dosimetric results were achieved: median PTV V100% = 94.5%; median PTV D98% = 89.9%; median lungs Dmean = 763 cGy; median left kidney Dmean = 1058 cGy; and median right kidney Dmean = 1051 cGy. For the plans prescribed to 1320 cGy total dose the following dosimetric results were achieved: median PTV V100% = 95.0%; median PTV D98% = 88.7%; median lungs Dmean = 798 cGy; median left kidney Dmean = 1059 cGy; and median right kidney Dmean = 1064 cGy. Maximum dose objective was met for all cases. The dose deviation between the treatment planning dose and the dose measured by the OSLDs was within ±4%. In summary, we have demonstrated that scripting can produce high‐quality plans based on predefined dose objectives and can decrease planning time by automatic target and optimization contours generation, plan creation, field and isocenter placement, and optimization objectives setup.


| INTRODUCTION
Total body irradiation (TBI) is a special radiation therapy (RT) procedure in which radiation is administered to the full body of the patient. In combination with chemotherapy, TBI is one of the therapeutic components of conditioning regimens used to condition patients with hematological neoplasms for hematopoietic stem cell transplantation (HCT), primarily those affected by acute myeloid leukemia (AML) and acute lymphoid leukemia (ALL). TBI enhances antineoplastic therapeutic efficacy due to its potential to reach sanctuary sites, such as the testis and the central nervous system (CNS), and provides immunosuppression that prevents bone marrow transplant rejection. [1][2][3][4] Radiation-induced interstitial pneumonitis is a major concern for patients undergoing TBI.
According to the International Lymphoma Radiation Oncology Group, pneumonitis occurs in about 25% of patients receiving fractionated TBI. 1 From a physics standpoint, guidelines for administering TBI are outlined in report no. 17 from the AAPM Task Group 29. 5 Historically, administration of TBI is delivered with the patient at an extended distance (extended SSD), such that the radiation field encompasses the patient's entire body. This technique continues as the standard of practice in most cancer centers performing TBI.
Open field treatments used for conventional TBI normally require the use of a beam spoiler to ensure coverage at shallow depths when using high-energy beams. Compensators are recommended as well in an effort to obtain homogenous dose distributions. 6,7 However, previous work has demonstrated that conventional hand calculations for TBI without tissue heterogeneity correction result in significant dose underestimation, particularly in the lungs for highenergy bilateral treatments. 8 One of the main drawbacks of at-distance open beam treatments used for TBI is the inability for selective sparing of organs at risk (OARs). Interstitial pneumonitis is the major side effect from highdose TBI radiation therapy and can be fatal in some instances. 1,[9][10][11] Additionally, high-dose TBI treatments can produce late toxicities, such as chronic kidney dysfunction and secondary malignancies. [12][13][14][15] In order to mitigate TBI-induced acute and chronic side effects, the use of shielding blocks, particularly for lungs and in some instances for kidneys, is widely accepted in high-dose TBI treatments. [16][17][18][19] However, accurate placement of beam modifying devices relative to the patient in treatment position presents challenges due to limitations in image verification, intrafraction patient motion, and reproducible patient setups.
More recently, several alternatives to at-distance treatment for TBI have been reported. Helical Tomotherapy (Accuray Inc, Sunnyvale, CA) has been employed in single institution studies for TBI and total marrow irradiation (TMI) treatments to obtain higher coverage for sites at high risk of recurrence while sparing major OARs, such as the lungs, liver, and kidneys. [20][21][22][23] In addition, there is ongoing effort to explore the use and feasibility of volumetric arc therapy (VMAT)-based TBI treatments in order to obtain a more homogeneous dose distribution, better target coverage, and better sparing of lungs, kidneys, and any other organs with increased risk due to patient comorbidities or previous radiation history. [24][25][26][27][28] Treatment planning approaches for VMAT-based TBI are not yet standardized, can be hard to develop, and may vary across institutions. Placement of fields, isocenters, and planning technique is solely based on each institution's individual efforts and experience due to the novelty of the technique and the lack of standardization.
The purpose of this work is to establish a fully automated approach for isocenter and treatment field placement as well as dose objectives selection for optimization for VMAT-based TBI treatment planning. This objective serves two purposes: First, to provide a consistent and standardized planning technique. Second, to ensure consistent shifts at treatment in order to prevent errors when treating multiple isocenters.

2.A | Subjects
Five full body anonymized CT scans from adult patients previously treated at our institution using conventional TBI were used retrospectively. CT images were obtained using a CT SOMATOM scanner (Siemens, Erlangen, Germany). Patients were simulated in the supine position. The protocol employed a 1 cm slice thickness, 500 mm acquisition diameter, and extended field of view (FOV) reconstruction of 650 mm. The use of anonymized retrospective CT scans for dosimetry studies was approved by our Internal Review Board and consent was waived. Additionally, an anthropomorphic body phantom (CIRS, Norfolk, VA) was scanned using the same CT simulator and a 5mm slice thickness. The phantom included all anatomical sections from mid-thigh to head with no upper or lower extremities.

2.B | Treatment planning
A script was developed to automate and standardize the treatment planning process. The script was developed in C# programming language using the Varian Eclipse v15.6 Scripting API (Varian Medical Systems, Palo Alto, CA). A summary of the treatment planning tasks that were automated is presented in Supplemental Figure S1. The input of the script is a structure set that must include at a minimum the following contours: body, lungs (including unilateral contours) and kidneys (including unilateral contours). These contours are created manually by the dosimetrist. Additionally, the user origin location needs to be entered prior to running the script. For these retrospective cases, an approximate location at body midline (anterior-posterior), lungs midline (craniocaudal), and sternum (left-right) was employed. For prospective patients to be treated with VMAT-TBI the user origin location will be determined at simulation and radiopaque ball bearings (bbs) will be placed to allow for user origin placement during treatment planning.
Script execution prompts the user to select a fractionation option and provide the intercept of the laser with the couch top longitudinal scale at the user origin location recorded at simulation (Fig. 1).
Currently, two options are available for planning: 150 cGy × 8 fractions BID to a total dose of 1200 cGy; and 165 cGy × 8 fractions BID to a total dose of 1320 cGy.
After the planner selects the option according to prescription, the script performs the following tasks automatically. The script creates one treatment course that contains two automatically generated plans, an upper body VMAT plan and a lower body 3D plan. The  Orfit couch tops and thermoplastic immobilization for VMAT-based TBI CT simulation. The presented thermoplastic devices were prepared on a healthy volunteer as part of establishing the program. Thermoplastic immobilization for the upper body board have fixed (one location) indexing. Thermoplastic immobilization for each individual foot can be placed in several locations in the board to allow for comfortable immobilization for patients with different heights as well as to control for leg separation.
• PTV is the body contour with a 5 mm margin inside the body surface and extended 3 mm into the lungs and kidneys.
• PTV_Sup is defined as the subsection of the PTV that is covered by the superior VMAT fields.
• PTV_Inf is the subsection of the PTV defined as the PTV minus PTV_Sup.
• PTV_Sup_Norm is defined as the PTV_Sup excluding the region of overlap with the inferior AP/PA fields.
Additionally, the script creates eight optimization structures.
Optimization structures are defined as the union of each individual VMAT field with the PTV structure. Due to the divergence of the fields, we consider the intercept at the coronal plane of the user origin. These optimization structures are named opt_ptv_X, where X is the beam number of each individual field from the upper body VMAT plan. Therefore, the optimization structures are named consecutively from opt_ptv_1 (union of the PTV and the area covered by first field of the head isocenter) to opt_ptv_8 (union of the PTV and the area covered by the second field of the pelvis isocenter).
The script will create all optimization objectives based on the fractionation selected and will load them automatically for optimization. Some additional options, such as aperture shape controller (set to moderate), air cavity correction (set to on), and jaw tracking (set to enabled), are set automatically by the script for the photon optimizer PO v.15.6. Finally, the script uses the user origin location, the couch intercept value entered by the planner in the graphic user interface, and the couch entered in the CT to provide the planner with all couch coordinates pertaining to the created isocenter arrangement. The script presents these values on the screen (Fig. 4), and creates a ".csv" file to be loaded in an in-house tool employed to assist with imaging and treatment delivery (out of the scope of this report). Additionally, the time required to run the script was compared to the time required to manually perform all the tasks completed by the script.
After script execution, the weights of the inferior 3D plan are manually adjusted to maximize coverage of the inferior target structure (PTV_Inf) while maintaining the maximum dose in that region below 130%. In the next planning step, a second instance of the script is executed to automatically run the optimization of the upper body VMAT plan using the preloaded dose objectives, and automatically calculate the plan after optimization. This second part of the script can be run in standalone mode, therefore, opening an independent instance of Eclipse, saving the progress and closing Eclipse allowing the planner to execute it after hours or overnight.
All treatment plans were created using a Varian TrueBeam machine with Millennium MLC, 2.5 mm optimization grid, 5 mm calculation grid, air cavity correction, and the analytic anisotropic algorithm (AAA). The energy employed was 6 MV photons for all fields with a maximum allowed nominal dose rate of 600 MU/min. F I G . 3. Initial VMAT field arrangement for two cases right after script execution. In both cases the total craniocaudal length covered by the fields remain the same (106 cm). Fields that share an isocenter overlap by 2 cm. Fields that not share an isocenter overlap by 5.3 cm. The final jaw shape is defined during optimization using jaw tracking. Left: Larger patient where maximum coverage of upper body VMAT fields stop mid-thigh. Right: Shorter patient demonstrating coverage up to the superior aspect of the knee. The superior jaw of the fourth field (chest isocenter, inferior field) in this case is automatically extended 3 cm to compensate for the isocenter location inferiorly in the lungs.
F I G . 4. Script output presenting treatment couch values to assist dosimetrist with treatment preparation. The script outputs each isocenter's longitudinal couch value and vertical couch values grouped by plan and field (in case an extended distance is required for the lower body plan) and lateral couch values for each plan. Additionally, these values are exported by the script into a ".csv" file.

2.C | Dose objectives
Two levels of planning goals were established for this treatment planning study. Primary goals must be met for every plan and include: target coverage of PTV V100% >90%, target near minimum dose of D98% >85%, and target maximum dose of PTV D2cc <130%. The primary goals for the organs at risk (OARs) include: lung mean dose <800 cGy for 1200 cGy plans or <900 cGy for 1320 cGy plans, individual kidney mean dose <1100 cGy, and maximum dose to any OAR D0.03cc <120%. Secondary goals include the following: target coverage PTV V100% >95%, target near minimum dose of PTV D98% >90%, and a target mean dose <110%. For the OARs, the secondary goal is lung mean dose <800 cGy for 1320 cGy plans.  All primary dosimetric goals were met for all calculated plans. A summary of the dosimetric results is presented in Table 1 for 1200 cGy total dose plans and in Table 2  are presented in Fig. 6 for both treatment plans (1200 cGy and 1320 cGy total dose) for one case.
The normalized dose deviation for all OSLD measurements was less than 4% for each individual OSLD. Individual values are presented in Table 3. Figure 7 illustrates the location of the OSLDs in the anthropomorphic phantom as well as the location taken in the treatment planning system to obtain treatment plan dose. Patientspecific QA for this plan obtained passing rate over 95% for each individual field and the areas of field overlap using a gamma criterion of 2% at 2 mm. The passing rate was >95% for all three modalities of patient-specific QA employed. Figure 8 presents the analysis using ArCheck and Portal Dosimetry for one of the chest isocenter fields.
Our plan uncertainty analysis revealed that the increase in the mean dose to the lungs was always below 3% for all plan uncertainties evaluated with the exception of a 10 mm lateral shift that resulted in a 4.3% mean lung dose increase. Regarding PTV coverage (V100%), our plan uncertainty analysis revealed that the decrease in coverage was below 3% for all scenarios analyzed. After analyzing dose profiles at field junctions we found a dose increase/decrease of about 20% per cm when uncertainty shifts were applied to a single isocenter in the longitudinal direction. This amount was expected considering the use of the auto-feathering tool in Eclipse and the overlap distance of 5.3 cm. However, there is some variability in areas mostly located at higher dose heterogeneity locations or around high/low-dose interfaces. Figure 9 includes an example of auto-feathering at one of the dose junctions.

| DISCUSSION
Our results demonstrate that automating VMAT-based treatment planning for TBI can reduce treatment planning time, increase consistency in isocenter placement, and decrease variability on field  do not produce large dose deviations as long as they are maintained within 10 mm. The largest uncertainties will occur if the planned longitudinal distance between isocenters is not maintained at treatment.
For this reason, we have developed an in-house software to assist with our image-guided radiation therapy (IGRT) approach. While a full description of this tool is out of the scope of this report, in brief, the software will prompt therapists to acquire IGRT at three locations (head, chest, and pelvis isocenters) and will calculate an optimal global shift based on the desired shifts at each location. Using this approach, we guarantee that the distance between all isocenters remains constant. The software calculates the residuals between the global shift and the individual shifts desired at each location and will This may actually be necessary if a particular case with a challenging anatomy is presented or if coverage/sparing in a particular area is desired. In any case, a workflow such as the one presented in this study should support standardization and robustness, and provide a good starting point. Additionally, an automatic tool like the one presented here has the potential to include options for additional OARs sparing or simultaneous integrated boost (SIB) regimens beside the two current regimens. There is a clear interest in the radiation oncology community to provide solutions to some of the common problems of TBI, such as organ sparing, lack of or limited imaging capabilities, dosimetric uncertainties, the need to manufacture compensators/blocks and patient comfort. Gruen et al. 22 reported their initial use of Tomotherapy for VMAT-based TBI treatments on ten patients treated to 12 Gy with 2 Gy per fraction. The lungs mean dose for this series was 9.14 Gy and no grade 3-4 toxicities were observed. Springer et al. 26 reported the use of VMAT-based TBI on a linear accelerator on seven patients. None of the patients reported severe lung toxicities and the authors were able to decrease the dose to the kidneys for patients with renal comorbidities to 7-8 Gy.
Tas et al. 25  MLCs cannot reach the end of the opposite bank due to larger field sizes. Two fields per isocenter with one field covering superiorly to the isocenter and a second one covering inferiorly (with a 2 cm overlap) proved useful to create MLC-based island blocks for the T A B L E 3 Normalized dose deviation between treatment plan system dose and OLSD measured dose. Differences were normalized using the prescribed dose (165 cGy per fraction). lungs and the kidneys. The script is able to adjust that overlap for the chest isocenter fields based on the distance between the chest isocenter and the user origin (Fig. 3) Our study has several limitations. First, the CT datasets employed were acquired on a different position compared to the one expected for a real VMAT-based TBI program (currently under development at our institution). In the CT datasets used the patient was positioned with forearms crossed over the chest, arms at the lungs level, and lower extremity separation similar to pelvis or shoulder separation. This positioning differs from our defined CT simulation positioning for VMAT-TBI with legs tighter together, arms at patient side and thermoplastic immobilization. While this is a limitation, our scripted treatment planning approach was able to provide high-quality plans even using the less favorable positioning for VMAT planning. Second, while the five cases explored here include patients with diverse body habitus, this is a very limited number and patients with even larger anatomic variation might pose a challenge at presentation. Finally, some information that the script uses for the field arrangement is specific to our institution. As an example, the script will always maximize the body region covered by the upper body VMAT plan. The longitudinal (craniocaudal) distance that is allowed is based on the indexing of the immobilization device to be employed at our institution and the longitudinal travel limit of the couch (with a margin to prevent reaching maximum travel). However, this does not conflict with the applicability of our field arrangement to other immobilization or setup scenarios.
F I G . 9. Dose profile at an area of field overlap between chest and abdomen demonstrating linear and smooth feathering between the two fields contributing to the total dose in the area with no abrupt change in field weight within the overlap region.
In our study, we have demonstrated that VMAT-based TBI treatment planning can be automated using scripting. Scripting can produce high-quality plans based on predefined dose objectives and can decrease planning time by automatic target and optimization contours generation, plan creation, field and isocenter generation, and optimization objectives setup. Additionally, a robust and standardized planning approach that accounts for couch longitudinal limits and immobilization facilitates treatment delivery ensuring consistent shifts and isocenter placement. Health) for their support developing the treatment planning approach and testing the treatment planning script.

CONF LICT OF I NTEREST
Honorarium from Varian Medical Systems (Jose Teruel).

AUTHOR CONTRIBU TI ON
Jose R. Teruel was involved in conception and design of the work, acquisition of data, data analysis, interpretation of data for the work, drafting the work, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.
Sameer Taneja was involved in design of the work, acquisition of data, interpretation of data for the work, data analysis, draft revision, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.
Paulina E. Galavis was involved in data analysis, interpretation of data for the work, draft revision, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.
K. Sunshine Osterman, Allison McCarthy, Martha Malin, and Naamit K. Gerber were in involved in interpretation of data for the work, draft revision, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.
Christine Hitchen was involved in conception and design of the work, interpretation of data for the work, drafting the work, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.
David L. Barbee was involved in conception and design of the work, acquisition of data, interpretation of data for the work, drafting the work, final approval of the version to be submitted, and responsible of accuracy and integrity of the work.