A clinical validation of the MR‐compatible Delta4 QA system in a 0.35 tesla MR linear accelerator

Abstract Purpose To validate an MR‐compatible version of the ScandiDos Delta4 Phantom+ on a 0.35T MR guided linear accelerator (MR‐Linac) system and to determine the effect of plan complexity on the measurement results. Methods/Materials 36 clinical treatment plans originally delivered on a 0.35T MR linac system were re‐planned on the Delta4 Phantom+ MR geometry following our clinical quality assurance (QA) protocol. The QA plans were then measured using the Delta4 Phantom+ MR and the global gamma pass rates were compared to previous results measured using a Sun Nuclear ArcCHECK‐MR. Both 3%/3mm and 2%/2mm global gamma pass rates with a 20% dose threshold were recorded and compared. Plan complexity was quantified for each clinical plan investigated using 24 different plan metrics and each metric’s correlation with the overall 2%/2mm global gamma pass rate was investigated using Pearson correlation coefficients. Results Both systems demonstrated comparable levels of gamma pass rates at both the 3%/3mm and 2%/2mm level for all plan complexity metrics. Nine plan metrics including area, number of active MLCs, perimeter, edge metric, leaf segment variability, complete irradiation area outline, irregularity, leaf travel index, and unique opening index were moderately (|r| > 0.5) correlated with the Delta4 2%/2mm global gamma pass rates whereas those same metrics had weak correlation with the ArcCHECK‐MR pass rates. Only the perimeter to area ratio and small aperture score (20 mm) metrics showed moderate correlation with the ArcCHECK‐MR gamma pass rates. Conclusions The MR‐compatible version of the ScandiDos Delta4 Phantom+ MR has been validated for clinical use on a 0.35T MR‐Linac with results being comparable to an ArcCHECK‐MR system in use clinically for almost five years. Most plan complexity metrics did not correlate with lower 2%/2mm gamma pass rates using the ArcCHECK‐MR but several metrics were found to be moderately correlated with lower 2%/2mm global gamma pass rates for the Delta4 Phantom+ MR.


| INTRODUCTION
Intensity modulated radiation therapy (IMRT) is a highly conformal external beam cancer treatment technique. Various factors such as commissioning data, dose calculation algorithm, delivery process, and performance of treatment delivery components have contributed to uncertainties in radiation therapy treatment. Patient-specific quality assurance (QA) for IMRT plans is often mandatory 1 and can be time consuming and laborious. As integrated linear accelerator and magnetic resonance imaging (MRI) hybrid system like Unity from Elekta (Elekta, Stockholm, Sweden) and MRIdian® from ViewRay® (View-Ray Inc, Cleveland, OH) continue to increase in popularity, QA solutions which are both MR-compatible and appropriate for IMRT also become necessary. The use of IMRT QA tools such as point detectors, arrays, film, etc. are well documented in the literature. 2 These same tools, however, which were previously incompatible with MRI are now being updated for use with MR-linacs.
When considering IMRT QA solutions the desirable properties of measurement devices include high resolution, the ability to calculate dose distributions in three dimensions, easy setup, wide clinical applicability (i.e., not tailored to only one delivery modality), negligible angular and energy dependence, a large active region, and the ability to be calibrated for and measure absolute dose. Although there is much data regarding the performance of IMRT QA detectors in conventional linear accelerators (linacs), the same cannot be said of MR-compatible versions of many of those same detectors. The presence of a magnetic field can impact the measurements from IMRT QA devices designed for use with conventional linear accelerators, 3,4 so MR-compatible radiation detectors are not as ubiquitous as conventional radiation detectors. In addition to ionization chamber specific studies, other groups have investigated the use of radiographic film, 4 cylindrical diode arrays 4,5 (i.e., ArcCHECK-MR), planar diode arrays, 6,7 GAFChromic™ EBT3 film, 8 gel dosimetry 9 for IMRT QA on MR linac systems. The ScandiDos Delta 4 , has been commercially available since 2015 and existing literature describe its characterization and commissioning 10  phantom. Finally, a series of fluence descriptive modulation complexity metrics were also calculated for each plan investigated in this work to determine which metrics, if any, might impact IMRT QA failure rates for both devices.

| MATERIALS AND METHODS
The MRIdian Linac system from ViewRay consists of a 6 MV flattening filter free linear accelerator sandwiched between 0.35 Tesla split superconducting magnet with a 28 cm gap between the two magnets. 13 Its MLCs are double stacked at the isocenter plane. The minimum and maximum programmable field sizes are 0.2 × 0.415 cm 2 and 27.4 × 24.1 cm 2 respectively. The MRIdian system uses a stepand-shoot intensity modulated radiation therapy technique to deliver dose that is calculated with a Monte Carlo algorithm. 13 The Sun Nuclear ArcCHECK-MR phantom is comprised of 1,386 diode detectors arranged in a helical pattern and spaced 10 mm apart around a cylindrical water-equivalent body. The length and diameter of the phantom are both 21 cm. The phantom also contains a cavity 15 cm in diameter which can accept various tissue equivalent inserts.
The ArcCHECK-MR used in this work included a relative calibration performed by the manufacturer and the absolute calibration was performed at our institution using a NIST-traceable A1SL scanning chamber (Standard Imaging, Middleton, WI) in a water tank at a calibration depth representative of the ArcCHECK-MR geometry. A 9.96 cm × 9.96 cm field size was used to irradiate the chamber at a depth of 3.3 cm (to match inherent ArcCHECK-MR buildup) in water at a source to surface (SSD) distance of 86.3 cm. 200 MU were delivered to the ionization chamber in this configuration and the chamber response was fully corrected for the effects of temperature, pressure, electrometer response, recombination, polarity, beam quality, and beam output in order to establish a known dose under reference conditions. The ArcCHECK-MR phantom was connected via  A total of 36 clinical plans were used to assess the performance of the device. These plans spanned several anatomical treatment sites previously treated on the ViewRay MRIdian linac including abdomen (13), lung (7), liver (9), and kidney (7). The abdomen group was comprised of abdominal sarcoma plans as well as pancreas treatments. At the University of Wisconsin, the ArcCHECK-MR phantom has historically been used for patient-specific IMRT QA of all clinical plans. Among the QA plans selected for this work were several with global gamma pass rates of less than 97% at 3%/3mm with a 20% dose threshold when previously measured on our Arc-CHECK-MR system. Those plans were of special interest since most clinical plans on the ArcCHECK-MR pass the 3%/3mm global gamma test at levels of 99% or higher. All QA plans were generated by transferring clinical plans to the ArcCHECK-MR geometry and recalculating them in the ViewRay treatment planning system (TPS). In the TPS, the magnetic field was set to ON, a grid resolution of 0.2 cm was used, and uncertainty during dose predication was set to 0.5%. Delta 4 Phantom+ MR QA plans in this work were generated in an identical manner, and in both cases, the geometric center of each phantom was placed into regions of high dose and low gradient to ensure that the high dose target region was being sufficiently sampled by the physical measurement.
Each QA plan was then delivered to the Delta 4 Phantom+ MR system. Dose differences at levels of 2% and 3% along with distance to agreement (DTA) values of 2 mm and 3 mm were recorded as well as 2%/2mm and 3%/3mm global gamma pass rates using a 20% dose threshold. These results were then compared to previous patient-specific QA measurements conducted at our clinic using the ArcCHECK-MR system.
For each treatment plan investigated, 24 modulation complexity metrics were calculated to determine whether any specific metric or metrics correlated with the gamma pass rates. The metrics and means of calculation are summarized in Table 2  Because the metrics investigated are primarily concerned with the shape of each segment's fluence, this treatment of ViewRay MLCs was thought to be more appropriate than calculating complexity metrics for each physical set of MLCs separately. Figure 1 illustrates an example ViewRay aperture and how the resulting composite shape, as opposed to individual leaf banks, are analyzed in this work.
The 2%/2mm gamma pass rates for both devices were then investigated as a function of the various complexity metrics to see which metrics, if any, correlated with the overall pass rate. This   Tables 3 and 4 summarize the 2%/2mm and 3%/3mm gamma pass rate statistics respectively. work, any p-value less than 0.05 was assumed to be statistically significant. All metrics which were at least moderately correlated with the gamma pass rates and involved an associated p-value of less than 0.05 are highlighted in bold in Table 5. Phantom+ MR plotted as a function of the irregularity metric for both radiation detectors. Figure 4 shows the same plotted data as Fig. 3 but distinguishes each data point based on the anatomical site of treatment. By further classifying the data into these anatomical groups, it became possible to see which specific plans contributed the most to the correlation between a given metric and the 2%/2mm global gamma pass rates. For example, Fig. 4b shows that the liver and lung data largely drive the negative correlation seen between the Delta 4 Phan-tom+ MR 2%/2mm global gamma pass rates and the irregularity metric. Conversely, the kidney plans measured using the Delta 4

| RESULTS
Phantom+ MR had almost no sensitivity to the irregularity metric (but also exhibited a noticeably narrower range of plan irregularity) and the abdomen data showed weak correlation with plan irregularity. plans, lower pass rates in 6 plans, and an identical pass rate in 1 plan. Figure 2a illustrates that the distinction of having "higher" 3%/ 3mm global gamma pass rates is largely irrelevant since both devices measured results well above our institutional criteria and that the measured 3%/3mm pass rates were similar for both devices, except for the specific ArcCHECK-MR low pass rate plans deliberately chosen from historical IMRT QA data. When considering the 2%/2mm global gamma pass rates, the Delta 4 Phantom+ MR measured higher pass rates in 24 plans and lower pass rates in 12 plans than the Arc-CHECK-MR device. Here, more pronounced differences were observed between the detectors. Importantly, the authors strongly emphasize that favoring a detector based solely on which one measures a higher gamma pass rate is a fundamentally flawed perspective and is not a suitable "apples-to-apples" comparison. The A corollary to not solely relying on global gamma pass rates to assess an IMRT QA radiation detector, however, is the desire to understand why potential differences could arise from different detectors in ostensibly similar circumstances and what to make of them. Work is presently ongoing with ScandiDos to attempt to establish detector-specific sensitivities to more clearly explain some T A B L E 5 Summary of Pearson correlation coefficients with the 2%/ 2mm global gamma pass rate for both the ArcCHECK-MR and the Delta 4 Phantom+ MR. Correlation values with a magnitude larger than 0.5 and an associated p-value of < 0.05 are highlighted in bold.

Metric All Plans
ArcCHECK-MR   Phantom+ MR gamma pass rates demonstrating moderate correlation with several plan complexity metrics. The ArcCHECK-MR 2%/ 2mm global gamma results appear to be largely independent of the complexity of the radiation treatment plan.