Robustness of single-isocenter multiple-metastasis stereotactic radiosurgery end-to-end testing across institutions

The accuracy of stereotactic radiosurgery (SRS) to multiple metastases with a single-isocenter using high definition dynamic radiosurgery (HDRS) was evaluated across institutions. An SRS plan was delivered at six HDRS-capable institutions to an anthropomorphic phantom consisting of point, film, and 3D-gel dosimeters. Direct dose comparison and gamma analysis were used to evaluate the accuracy. Point measurements averaged across institutions were within 1.2±0.5%. The average gamma passing rate in the film was 96.6±2.2% (3%/2 mm). For targets within 4 cm of the isocenter, the 3D dosimetric gel gamma passing rate averaged across institutions was >90% (3%/2 mm). The targeting accuracy of high definition dynamic radiosurgery assessed by geometrical offset of the center of dose distributions across multiple institutions in this study was within 1 mm for targets within 4 cm of isocenter. Across variations in clinical practice, comparable dosimetry and localization is possible with this treatment planning and delivery technique.


INTRODUCTION
For the treatment of intracranial brain metastases, the accuracy and efficiency of stereotactic radiosurgery (SRS) to treat multiple targets concurrently has been previously described in the literature ( . Specifically, the ability of Elekta Versa HD™ (Stockholm, Sweden) equipped with the Agility™ MLC, HexaPOD™ evo RT six-degree-of-freedom tabletop and in conjunction with the Monaco® treatment planning system (TPS) was investigated in this study at multiple institutions to show reproducible accuracy across variations in practice. The combined use of these technologies with cone-beam CT image guidance and six-degree-offreedom couch patient positioning (enabled by XVI and HexaPOD) is termed by Elekta as high definition dynamic radiosurgery (HDRS). HDRS is characterized by the integration of optimization, planning, imaging and dose delivery techniques. Details of this technique and its accuracy were previously published in 2018 (Saenz et al, 2018). The TPS is Monte Carlobased and takes advantage of Agility MLC capabilities to effectively reduce the 5 mm leaf width in the jaw direction by positioning the Y-jaws within leaf widths on either end of the target(s). Since stereotactic targets may only require a few open leaf pairs, this jaw positioning accuracy can significantly improve target conformality. Treating off of the central-axis requires continuously travelling MLCs to track the targets as the gantry rotates, which is accomplished by MLCs travelling up to 6.5 cm/s. High geometric and dosimetric certainty is required for radiosurgery, and uncertainties arise from many sources. Hence, end-to-end tests are critical to examine the localization and dosimetric accuracy of the entire process (Halvorsen et al, 2017). Our previous work (Saenz, 2018) described how HDRS dose distributions were validated at our institution with a patient-specific phantom, tailored to a particular patient's anatomy and filled with a dosimetric gel, in conjunction with similar phantoms equipped with point dosimeter and film inserts. However, such a technique is subject to differences in clinical practice between institutions due to variations in simulation technique, treatment planning strategy, alignment process and other factors. The measurement protocol requires standardization to obtain consistent results when multiple institutions are involved.
This validation project was designed at the Mays Cancer Center at UT Health San Antonio as described in the previous publication. Other institutions included in the study are Ludwig Maximilian University of Munich (Munich, Germany), Farrer Park Hospital (Farrer Park, Singapore), Southeast Health (Cape Girardeau, Missouri, USA), Rambam Institution (Haifa, Israel), Franciscan Health (Indianapolis, Indiana, USA) and the German Oncology Center (Limassol, Cyprus). All institutions utilized the Elekta Agility collimator HDRS linear accelerator (Versa HD or Infinity) in conjunction with the Monaco treatment planning system. Results from all institutions were pooled and analyzed together for assessment with the exception of one site which did not have a HexaPOD six-degree-of-freedom tabletop, which is crucial for the overall accuracy of radiosurgery (Dhabaan et al, 2012). Hence, results for HDRS are best described by the six institutions with this capability.

Treatment Planning Dataset
An existing RT structure set from a previouslytreated radiosurgery patient was chosen for the purpose of this study. The six metastatic lesions were modified in size in the CT data set (0.74 × 0.74 × 1.25 mm voxels) so that a range of targets with diameters from 6 -25 mm could be treated. In addition, a larger seventh target was included for dose normalization purposes. Target locations included lesions sufficiently far apart to measure up to 5.6 cm from isocenter to test the effects of rotational uncertainties on localization accuracy across the brain. A list of targets including size and location are shown in Table 1. Targets were located in the right parietal, left parietal, left frontal, right occipital, and brainstem regions.

Phantom Design
The endpoint of this study was to assess accurate localization and dosimetry in an anatomically realistic measurement. Therefore, it was crucial to perform 3D dosimetry, which is possible with a gel dosimeter. The RTsafe PseudoPatient™ gel phantom produced by RTsafe P.C., (Athens, Greece) was used in this study as has been described in the literature by Kalaitzakis et al (2016) and Maris et al (2016). This dosimeter was chosen because it is a 3D dosimeter that can be cast in nearly any form, allowing for measurement in a patientspecific geometry. A modified composition of VIPAR polymer gels was used constituting of a mixture of the monomer N-vinylpyrrolidone in 6% weight fraction (wf), the cross-linker N,N'-methylene-bisacrylamide in 4% wf, gelatine of type A in 5% wf, deionized water in 85% wf, and 7 mM THPC (Saenz et al, 2018, Pappas et al, 1999, Papadakis et al, 2007. The primary advantage of gel dosimetry in an anthropomorphic phantom is that, unlike other patient-specific QA, it does not rely on a recalculation of the plan on a phantom nor on the process with which to reconstruct a 3D dose distribution. Rather, the measurement in this phantom can be directly compared with the patient's calculated dose distribution as was demonstrated by Makris et al (2019) who have shown that the phantom is dosimetrically and anatomically equivalent. Finally, for the most realistic execution of an end-to-end test, an anthropomorphic phantom is the preferred tool due to its direct representation of patient anatomy.
For the purposes of this study, three phantoms were produced by RTsafe based on the actual CT data set bony anatomy and external contour of our reference patient. The first was filled with the aforementioned dosimetric gel so that 3D dose measurements could be obtained. In addition, the other two phantoms were modified to accommodate placement of the other detectors. One had an insert for one of two point dosimeters: Standard Imaging (Middleton, WI, USA) Exradin A16 or A26 ionization chamber (inside diameters of 2.4 and 3.3 mm respectively with collecting volumes of 0.007 and 0.015 cm 3 respectively) or PTW (Freiburg, Germany) microdiamond detector (sensitive volume of 0.004 mm 3 ). The other was designed with a holder for a Gafchromic film (Ashland, Bridgewater, NJ, USA). The point dosimeter and film cassette were situated to coincide with the center of the larger dose normalization target (3 cm from isocenter). By making point dosimeter measurements in a larger target less sensitive to small field dosimetry complications, the absolute dose accuracy can be shown with the point and film dosimetry and relative dose distribution characterized by gel dosimetry.
A planning template for HDRS delivery was devised in the Monaco version 5.11.02 treatment planning system (Elekta, Stockholm, Sweden) at UT Health San Antonio. This template consists of a set of beam geometries, prescription, and optimization objectives to be used across all institutions. The isocenter was set at the centroid of the targets. Five non-coplanar VMAT arcs were set up in the template and Table 2 demonstrates the arc parameters. As the dose linearity of the gel dosimeter breaks down above 12 Gy, the targets were assigned a prescription dose of 8 Gy with maximum doses remaining under 12 Gy. In all metastases, the dose was allowed to peak up to 12 Gy, while a homogeneous dose of 8 Gy was planned in the dose normalization target. No normal tissue objectives were defined and instead conformality at 8 Gy and 4 Gy was pursued. Target penalty, quadratic overdose and conformality objectives were used for optimization. A 1 mm dose calculation grid spacing was used with 1% statistical uncertainty per calculation of dose to medium. In the five single arcs, 180 control points were allowed with 0.5 cm minimum segment width. Medium fluence smoothing was used. For the multi-institutional study, the CT data set, RT structures and Monaco planning template, as well as the phantoms encompassing the point dosimeter and the Gafchromic cassette insert, were shared between all institutions in the study for the purposes of consistency of application. Upon receiving this data, institutions created their own treatment plans following the planning template. Therefore, the beam geometries, prescription, and optimization objectives are equivalent between all institutions. However, each institutions plan has its own MLC plan and monitor units.

Treatment Delivery
Patient data was sent to the Versa HD linear accelerator for delivery. The gel phantom was treated first and was localized with an XVI VolumeView™ cone-beam CT scan, which was fused with the reference CT data set. The grey value (T + R) (translation and rotation) algorithm in XVI was used with a clipbox placed such that the entire brain was covered. The correction reference was placed at the isocenter. 6D corrections were applied using the HexaPOD tabletop prior to delivery. The process was repeated for the point dosimeter and film measurements.

Film dosimetry and uncertainty
Gafchromic EBT3 film was used for the film phantom measurement. The 14 x 8 cm 2 Gafchromic EBT3 films were calibrated at the Secondary Standard Dosimetry Lab of the Greek Atomic Energy Com- Film was shipped to each institution and evaluated upon return providing dose maps in absolute Gy values. Since it can take varying durations for an institution to perform the irradiations, the time-dependent calibration curve was built using films scanned at 24 and 72 hours, as well as 5 and 8 days after irradiation. An EPSON (Seiko Epson Corp., Japan) V850 PRO flatbed scanner was used with 150 dpi digitization and 48-bit depth. More details are presented in our previous work on end-to-end accuracy at a single institution (Saenz et al, 2018).
The main source of the uncertainties involved in EBT3 film measurements is the calibration procedure and specifically the calibration curve fitting process (Aldelaijan et al, 2011). An estimation of the range of the combined 1σ dosimetric uncertainties was performed by following the methodology described in

Dose extraction from Gel phantom
The gel phantoms were scanned on the MRI units of each institution one day after irradiation. Gel dose read-outs were performed using a variety of MRI units including, GE 1.5T Sigma HDX and 3T Optima 750 MW, Siemens 1.5T Espree and Sonata Vision and 3T Trio Tim, as well as Phillips 1.5T Ingenia. A 2D, multi-slice, multi-echo, Half Fourier Single Shot Turbo Spin Echo (HASTE) proton density to T2-weighted sequence was implemented sequentially using the head coil. Averaging was set to n = 14 to improve signal-to-noise ratio. MR-related geometric distortion was reduced with a bandwidth set to 1220 Hz/pixel. The MR protocol used is the one developed and recommended by the gel producer (RTsafe P.C.). Since the relaxation rate (R2 = 1/T2) of the polymer gel is directly proportional to absorbed dose, the MR scan was compared with the patient-derived planning CT data set for dose analysis. The acquired MR images were spatially co-registered with the corresponding CT datasets by performing bone-rigid registration using Monaco since the phantoms provide adequate bone and soft tissue signal in both CT and MR images. Doses were normalized at each site to an appropriate ROI in the larger dose normalization PTV. Each institution performed a CT scan of the phantom with the ionization chamber or the microdiamond detector in place in order to determine the sensitive volume of the used point detector and delineate the corresponding structure for dose calculation and comparison purposes. 3D gamma analysis and dose profile comparisons were performed for analysis.
The linear relationship between measured R2 relaxation rates and delivered dose up to 12 Gy has been verified in previous studies including Pappas et al (1999) and Pappas et al (2019) for the same gel formulation used in this work and for a time window from 24 hours up to 15 days between gel irradiation and MRI scanning (Pappas et al, 1999, Papoutsaki et al, 2013.The gel dose uncertainty component, regarding the duration between the gel fabrication and irradiation for a time interval of 6 days has been found less than 2% 18 . For the purposes of this study, only relative gel dose measurements were performed in order to avoid any potential sources of uncertainty that affect absolute gel dosimetry (Baldock et al, 2010). Uncertainty level from the fitting procedure (linear fit) was calculated based on Papadakis et al (2007) methodology. An estimation of the main sources of uncertainties for relative dose measurements was performed, revealing a dosimetric 1σ uncertainty less than 3%. Moreover, taking into account the spatial resolution of the MR images and the registration technique between MR gel data and CT dataset of the gel-filled phantom, a spatial 1σ uncertainty less than 2mm.

Data analysis (phantom evaluation)
Point dosimeter measurements were evaluated by comparing to the average dose in a region of interest (ROI) indicating the position of a point dosimeter's sensitive volume in the plan calculation. Each institution followed standard dosimetry protocols in the conversion of measured charge to absorbed dose using k q and other correction factors from a previous machine calibration. The point dosimetry measurement established the absolute dose levels for the gel dosimetry. Measured 2D and 3D doses were compared to the planned dose distribution after following a rigid image registration procedure using the film fiducials for the film analysis and using the cranial anatomy for the MRI scan. From this point, direct dose profile comparisons can be made. 3D gamma analysis for both the film and gel dosimeter was conducted by utilizing global gamma analysis with various passing criteria. Gamma analysis was performed in the film plane ROI and in the region of individual targets for the 3D gel analysis using the DICOM RTStruct file. In both dosimetric systems, the measured dose distribution was set as the reference one and the TPS calculated as the evaluated during gamma calculations applying a dose threshold of 20% to avoid voxels with increased uncertainties since in both dosimetric methods the uncertainties are dose dependent. The Euclidian distance between the center-of-mass of the calculated and measured dose distributions were also evaluated as a metric of spatial agreement. This metric has been presented and established previously (Moutsatsos et al, 2013 andPantelis et al, 2018).

Point dosimeter measurements
Point measurements made in the dose normalization target agreed with TPS calculations (mean dose within a contour of the detector) The percentage difference between planned and measured dose was calculated (Table 3). A maximum percent difference of 1.7% was observed. The mean percent difference was 1.2% with a standard deviation of 0.5%.

Film Analysis
3D Gamma analysis was used to analyze the measured dose distribution in the plane of the film insert. This means that the measured dose plane was compared with the 3D dose distribution from the treatment planning system, not just a planar dose. At 3%/2 mm and 2%/2 mm, the average and standard deviation of gamma passing rates were 96.9 ± 2.1% and 95.1 ± 2.8% respectively. A visual indication of the gamma results in the film plane is shown in Figure 1 along with the tabulated results in Table 4.

Gel Phantom Analysis
For the 3D gel phantom analysis, gamma distributions within the targets were analyzed using the dose information from the T2 values and the planned 3D dose distribution. This was analyzed in conjunction with target diameter and distance from isocenter. Figure 2 shows example isodose lines from measurement and calculation in the indicated region of interest. At 3%/2 mm, the average gamma passing rate across institutions was ≥ 90% for targets within 4 cm of isocenter. Two targets were beyond 4 cm of isocenter, for which an average gamma passing rate of 82.7% and 93.3% was measured. Table 5 summarizes this data.
An illustration of measured dose accuracy repeatability across institutions is shown in Figure 3 where dose profiles for each site are plotted together. Figure 3 also illustrates the physical positioning accuracies found for all targets and all institutions quantified by assessing the center-of-mass of the dose distribution in the measured and calculated 3D data and reporting the distance between them. It is plotted in the order of target volume as well. In this study, < 1 mm spatial accuracy for targets within 4 cm of isocenter was found. Beyond 4 cm, the highest spatial deviation recorded was 1.9 mm.

DISCUSSION
Agreement between HDRS measurement and TPS calculation was repeatable across institutions in this study where we are extending targeting accuracy from a point (conventional SRS) to an 8 cm diameter sphere. This was verified using 3D gamma analysis for both film and gel phantom measurements. Absolute dose verification within 2% is well within recommendations for SRS end-to-end dosimetric accuracy (5% per AAPM MPPG 9.a), demonstrating satisfactory results when compared with conventional SRS (Halvorsen et  al, 2017).
The advantages and critical need for the use of 6D corrections was re-affirmed by this study. The crossconsortium mean gamma passing rate for the targets beyond 4 cm was raised by 5.0% and 6.5% simply by excluding results from one site without HexaPOD rotational corrections. 6D corrections are essential for the end-to-end accuracy of a multiple-target SRS program since small rotational errors are of an increasing magnitude with increasing distance from isocenter and with decreasing target diameter. Even a 1-degree rotation can result in a 1 mm positioning error for a target 6 cm beyond isocenter. Particularly for small targets at this distance, rotational errors can result in substantial geometric miss unless rotational corrections are performed. Of note, rotational uncertainties arise from multiple sources (patient positioning, collimator angle, and couch angle when applying couch rotations without fiducials) and are all inherent in this study. Rotational uncertainties remain a concern for a stereotactic delivery even with 6D corrections, although they are much smaller. When analyzing dose profiles across the targets, many profiles show excellent spatial agreement while others show spatial offsets generally with the 1 mm physical positioning accuracy. While some slight localization error is expected, due not only to inherent uncertainties in the technique but also due to uncertainties in the gel phantom analysis, it remains prudent to understand the applicability and limitations of this technique. In regions distant to the isocenter, attention should be given to the clinical implications of geometric misses, particularly when lesions near organs at risk are involved. Typical SRS accuracy of <1 mm breaks down beyond this region. Nevertheless, clinically acceptable plans can be created with applied setup margins with careful attention paid to cumulative metrics such as dose conformity and gradient index for  End-to-end single isocenter SRS QA all targets and total volumes of normal brain receiving a safe dose level. Alternatively, it may be wise to use separate isocenters to treat two groups of targets. While this may reduce the efficiency of the technique, it represents a compromise between efficiency of delivery and accuracy of dose delivery. Of note, it was observed that T05 showed worse gamma results than the other targets. T05 is second smallest target as well as the second furthest from the isocenter. Hence we would expect these targets to be most sensitive to uncertainties. However, the results for T03 were even better than for T05. Upon further analysis, it was determined that the 8 Gy isodose distribution tended to be more forgiving for T03 than for T05 leading to the relatively good agreement between measurement and calculation for T03. This emphasizes the need for careful inspection of a treatment plan not only for ideal coverage, conformity index, and gradient index, but also with attention given to robustness analysis.
Being the first 3D end-to-end study of its kind (the first multi-center, multi-target accuracy study), the results of this study suggest the feasibility of establishing standards for HDRS accreditation or audits for newer institutions considering this technology. The quantitative results also can serve as a benchmark for assessing baseline accuracy and can aid in the commissioning process. Limitations of the study include need to use the same structures at each site. This was necessary to compare results across institutions but did limit the degree to which the study was a true end-to-end test at specific institutions. Furthermore, no verification imaging was conducted at couch angles other than the nominal angle. This could be resolved by use of inroom or gantry-mounted kV planar imaging or potentially with surface imaging (Yin et al, 2009).

CONCLUSIONS
Precise stereotactic radiosurgery to multiple targets in the brain with a single isocenter has been demonstrated across multiple institutions to be a feasible approach. The combination of Monaco, Versa HD with HDRS, Agility and HexaPOD enables successful treatment of multiple brain metastases using a single isocenter SRS technique and the end-to-end