Unlocking a closed system: dosimetric commissioning of a ring gantry linear accelerator in a multivendor environment

Abstract The Halcyon™ platform is self‐contained, combining a treatment planning (Eclipse) system TPS) with information management and radiation delivery components. The standard TPS beam model is configured and locked down by the vendor. A portal dosimetry‐based system for patient‐specific QA (PSQA) is also included. While ensuring consistency across the user base, this closed model may not be optimal for every department. We set out to commission independent TPS (RayStation 9B, RaySearch Laboratories) and PSQA (PerFraction, Sun Nuclear Corp.) systems for use with the Halcyon linac. The output factors and PDDs for very small fields (0.5 × 0.5 cm2) were collected to augment the standard Varian dataset. The MLC leaf‐end parameters were estimated based on the various static and dynamic tests with simple model fields and honed by minimizing the mean and standard deviation of dose difference between the ion chamber measurements and RayStation Monte Carlo calculations for 15 VMAT and IMRT test plans. Two chamber measurements were taken per plan, in the high (isocenter) and lower dose regions. The ratio of low to high doses ranged from 0.4 to 0.8. All percent dose differences were expressed relative to the local dose. The mean error was 0.0 ± 1.1% (TG119‐style confidence limit ± 2%). Gamma analysis with the helical diode array using the standard 3%Global/2mm criteria resulted in the average passing rate of 99.3 ± 0.5% (confidence limit 98.3%–100%). The average local dose error for all detectors across all plans was 0.2% ± 5.3%. The ion chamber results compared favorably with our recalculation with Eclipse and PerFraction, as well as with several published Eclipse reports. Dose distribution gamma analysis comparisons between RayStation and PerFraction with 2%Local/2mm criteria yielded an average passing rate of 98.5% ± 0.8% (confidence limit 96.9%–100%). It is feasible to use the Halcyon accelerator with independent planning and verification systems without sacrificing dosimetric accuracy.


| INTRODUCTION
The Halcyon™ radiotherapy platform (Varian Medical Systems, Palo Alto, CA, USA) is designed as a self-contained system, combining an integrated treatment planning system (TPS), information management/Record and Verify (R&V) system and a ring-gantry radiation delivery device. The beam model in the integrated TPS (Eclipse) is provided and locked down by the vendor. The machine is essentially tuned to match a set of predefined beam specifications, including the percentage depth doses (PDDs) and cross-beam profiles. The rest of the dosimetric commissioning process is simply validation of the noneditable TPS model. A portal dosimetry-based system for patient-specific dosimetric QA (PSQA) is also a part of the preconfigured package. The tests of the entire system and its components have been thoroughly described in the literature, for both prototype and clinical systems. [1][2][3][4] While the advantages of the preconfigured system in terms of uniformity across the user base are undeniable, this approach may not always be optimal in practice. Case in point is implementation in our large radiotherapy department with over 50 active TPS users. A new TPS has been recently installed and commissioned as the primary planning system for the majority the linear accelerators in the department. Providing an adequate number of planning licenses on a Halcyon-specific TPS, as well as training and maintenance, for a single accelerator would constitute a non-trivial financial and human resource burden. Consequently, we set out to explore if it was possible to leverage the existing TPS capabilities to plan for the Halcyon accelerator without compromising accuracy. In addition, an independent semiempirical system for PSQA was also validated for use with Halcyon since it is already in use throughout the department.

2.A | Treatment planning and delivery
The Halcyon (v. 2.0) is configured with a single 6 MV flattening filter free (FFF) beam (nominal dose rate 800 cGy/min at d max of 1.3 cm).
It is equipped with a stacked-and-staggered dual-layer MLC. 3 While each leaf casts a 1 cm shadow at isocenter (100 cm from the source), the staggered design results in effective leaf with of 0.5 cm.
The MLC is single focused (in-plane) and the leaves have rounded ends. The curvature radius (23.4 cm) is larger than for the standard (8 cm) or high-definition (16 cm) Varian 120-leaf MLCs. 3 Leaf height is 7.7 cm vs 6.5 to 6.75 cm for 120-leaf collimators (depending on the model and leaf position in the bank), reducing leaf transmission. 3 The maximum field size at isocenter is 28 × 28 cm 2 . Machine scales are fully compliant with International Electrotechnical Commission (IEC) Standard 61217.
Eclipse shares the database with Aria v. 15.6.7 R&V system (Varian) which in turn passes the information to the accelerator. For commissioning purposes, the R&V system can be bypassed in the Service mode and DICOM RT Plan files loaded directly through the accelerator console.
The primary TPS investigated in this study was RayStation v.9B (RaySearch Laboratories, Stockholm, Sweden). It fully supports the Halcyon geometry, including the double-stacked and staggered MLC configuration, and dose delivery parameters. As with the rest of our machines, we commissioned only the Monte Carlo (MC) dose calculation engine. For comparison to Eclipse, the identical plans were recalculated with the vendor-configured TPS supplied with the Halcyon system, using the grid-based Boltzmann equation solver (ACUROS™) 5 algorithm v. 15.6.06.

2.B | Experimental data collection 2.B.1 | The roadmap
The dosimetric commissioning process followed these general steps.
The cross-beam profiles and PDDs were collected in a water tank with different detectors and used for the beam energy and profile modeling. Relative output factors were collected with appropriate detectors to establish output correction factors in the beam model. Specials measurements were taken to help with MLC modeling.
Those included dynamic fields measurements with a Farmer chamber following the dosimetric leaf gap (DLG) concept 6,7 and various static abutting fields scanned with a small diode. Finally, a set of modulated plans was used to hone the model against multiple ion chamber measurements and additionally verify it with a diode array.

2.B.2 | Beam profiles and PDDs
The MC algorithm implementation in RayStation follows a mixed approach. The accelerator head is not simulated, but rather the phase space above the moving parts is deduced form the experimental dose distributions in water. 8 The resulting fluence is modulated by the transmission values of the MLC leaves (Halcyon has no movable jaws). The MC simulation starts downstream once the fluence encounters the patient boundary described by the external contour.
To establish the phase space, a set of beam profiles and PDDs was collected. In addition to verifying the machine-specific dose distribution, the goal was to compare the ion chamber and diode scans for the Halcyon beam and specific detectors, to optimize the commissioning process. The detailed description of the collected data and instrumentation is tabulated in the Appendix.
PDDs for the fields ≤10 × 10 cm 2 were indistinguishable between the Edge diode (Sun Nuclear Corp. Melbourne, FL, USA) and CC13 IC (IBA, Schwarzenbruck, Germany). For the fields larger than 10 × 10 cm 2 , cross-beam profiles obtained with the diode were used for MLC leaf end modeling while the IC data helped with adjusting the beam shape well within the field. The CC13 IC and Edge diagonal profiles for a 28 × 28 cm 2 field showed a small but consistently noticeable difference away from the central axis (about 1%).
For the step-and-integrate small field PDD measurements, the search for the maximum signal was performed at d max and every 5 cm in depth thereafter. The Halcyon MLC geometry does not allow for a centered 0.5 × 0.5 cm 2 field. However, it is possible to construct such a field with the center shifted 0.25 mm perpendicular to the leaf movement direction. The profiles and PDDs were intercompared between the standard beam data, our measurements with two detectors, and RayStation calculations. These comparisons included one-dimensional gamma analysis with the open source "MPPG 5a tool." 9

2.B.3 | Relative output factors
In RayStation, introduction of a relative output value for a particular field size is possible only if a corresponding depth-dose curve is also present. Collecting the small-field PDDs allowed us to include smallfield values in the output optimization process. Among the three detectors used the IC and water-equivalent scintillator (W1, Standard Imaging Inc., Middleton, WI, USA) require no field size corrections in the respective ranges of field sizes. 10 between the Edge (with those corrections) and scintillator measurements to within ≤1% for the square field sizes down to 0.415 cm on a side. The detector was centered in the field by searching for the maximum signal in two dimensions. The Edge and CC13 IC were horizontal, while the W1 was positioned with the long axis toward the source (i.e., 1 mm collecting volume diameter and 3 mm length).
The output data were compared against the RayStation calculations.
Here and elsewhere the MC calculation statistical uncertainty (one standard deviation averaged between all voxels receiving ≥50% of the maximum dose) was set to 0.3%. For the smallest fields (≤2 × 2 cm 2 ) the central voxel dose was taken as the RayStation output. The dose grid was 1 × 1 × 3 mm 3 . For the rest of the fields, an isotropic 2 mm grid was used and the mean dose to a small cylindrical region of interest approximating a CC13 chamber represented the calculated output. Leaf end (MLC offset). The data collected in this section included both static and dynamic measurements carried out at the central axis and shifted across the field in the leaf travel direction. 6 The field edge positions (as defined by the inflection point on the penumbra profile of an FFF beam 12 determined by the extremum of the first derivative of the penumbra profile) were extracted from the diode profiles to assess the linearity of MLC positioning across the field.

2.B.4 | Data for fine-tuning MLC parameters
Another static measurement was the abutting fields test, which is quite sensitive to the MLC end modeling parameters. Two abutting 2 × 4 cm 2 (IEC X × Y) fields were separately scanned in the X direction at the depth of 10 cm with the Edge detector. The apertures were defined by both leaf layers. Care was taken to always scan in the same direction to minimize the positional uncertainty. The scanning curves were normalized to the center of the respective openings and summed. The data from three runs were averaged. The summary profile was compared to the high-resolution calculations (1 mm 3 voxel) in RayStation. This procedure was performed with the abutment line at 0 and AE10 cm X positions.
Another approach to the leaf-end modeling is dynamic measurements involving MLC-defined gaps sweeping across the field. The

2.B.5 | MLC parameters in the RayStation model
Ideally, the algorithm should ray trace through the MLC modeled with correct dimensions and density to arrive at the fluence downstream, as demonstrated by Losasso et al in 1998. 7 However, the MLC model in RayStation employs significant simplifications. The leaf has no height but is rather treated as an infinitely thin object with user-specified transmission. 13 1) and (2) for the left and right MLC banks, respectively. 8,14 where X nom is the nominal leaf position and Offset, Gain, and Curvature are the MLC parameters in the RayStation Fluence tab. The X coordinate is positive for both left and right leaf banks to the right of the central axis and is negative to the left of it, as seen in beam's eye view.
The analysis of these equations is best carried out with introduction of two more variables. For a pair of opposing leaves both with the nominal planned position X nom , the dosimetric offset, The chambers' daily correction factors were obtained by cross-calibration to RayStation dose at isocenter in the phantom in the parallel-opposed 10 × 10 cm 2 fields. Whenever feasible, the VMAT/IMRT plans had additional objectives to make the dose in a small volume surrounding the chambers' locations as homogeneous as possible.
One chamber was placed in the high-dose region (at the isocenter) and another in the lower dose location. The low to high dose ratios ranged from 0.4 to 0.8. All percent dose errors were reported normalized to the local dose, which is an unbiased and stringent measure of the algorithm accuracy.

3.A | Beam profiles and PDDs
Representative cross-beam profiles and PDDs were compared to our water scans with gamma analysis using 2%L/1 mm criteria for the 2 × 2, 10 × 10 and 28 × 28 cm 2 fields. The cutoff thresholds were adjusted per field size (3 to 5%) to include the outside toe of the penumbra but exclude the noisy low-dose regions outside the pri-    However, for such a narrow field the relative output is not the same as for 0.5 × 0.5 cm 2 . The output was calculated outside of the beam editor for an asymmetric field and the output correction factor was adjusted manually. As one can see in Fig. 2, this procedure leads to correct results (the depth doses are presented in dose per MU). The collimator exchange effect for orthogonal elongated fields is minimal but the small difference is correctly represented by RayStation (arrows in Fig. 3). Note that due to the depth/SSD differences the output factors are numerically different from Eclipse.

3.C | MLC parameters
3.C.1 | Tongue-and-Groove width As evident from the combined complementary bars scans in Fig. 4 T&G width is expected to have at most a modest effect on the final dosimetry. 16 Also note the consistency in the Halcyon MLC gap width across the field and the ability of both the diode scan and calculation on a 1 mm grid to resolve the dose "bump" at the outer edge of the leaf resulting from the reduced thickness.

3.D | The MLC Gain parameter
The differences between the actual and nominal mid-point positions of the two opposing leaves planned for the same position X nom , or ΔX MP , are presented in Fig. 5 as a function of X nom . Analysis of the simple linear fit shows that the slope of the regression line is not statistically significantly different from zero (P = 0.07). 30 We attributed the observed variations to measurement error 14 and set the Gain value in the model at 0.0.

3.E | The Offset and Curvature parameters
The leaf offset is one of the most important parameters influencing the final IMRT/VMAT results. 16 From the static abutting field where a 0.01 cm change is readily apparent (Fig. 6). Even with the Millennium MLC, the leaf the tip width parameter in RayStation should have only a moderate effect on the resulting dosimetric agreement. 16 It is expected to be even smaller for a taller Halcyon leaf with a flatter end. Absent evidence to the contrary, the leaf tip width was fixed at 0.0 cm to minimize the number of variables.

3.G.1 | Ion chamber measurements
The first step was to optimize the Offset parameter to minimize the mean difference between the measured and calculated point doses.
The graph of the average dose difference across 30 measurements with three different planning techniques, as a function of the Offset value, is presented in Fig. 8. With the plans recalculated with Eclipse, the overall mean difference between the TPS and measured dose was 1.5% AE 2.0%. This average was clearly skewed by the large locally normalized deviations (~8%) in the low-dose high-gradient region (C-Shape  The follow-up Holm-Sidak's multiple comparisons test 30 showed that all pair-wise differences were also statistically significant (P ≤ 0.007).
In clinical practice, the RayStation calculated dose would normally be compared to the PF dose reconstruction by gamma analy- Combined with tight manufacturing specifications, such approach ensures that minimum quality standards are met across the user population and the first published results indicate encouraging results with the IROC end-to-end tests. 2,36 However this approach also has its drawbacks. While having deep knowledge of their system, the vendor team may not be exposed to the full variety of clinical scenarios. A set of TPS parameters optimal for one set of clinical plans may not necessarily be the best choice for other types. 31 A local physicist might add experimental data that would make the modeling dataset more robust. Finally, for various logistical reasons a clinic may find it beneficial to rely on a TPS of their choice to avoid the need for multiple accelerator-specific systems.
All of the above considerations were taken into account in our decision to commission an independent vendor TPS for the newly installed Halcyon accelerator.
We have made the beam dataset used for modeling more complete, by adding dose profiles scanned with both a diode and an ion chamber, and also provided relative output factors for very small fields, down to 0.5 × 0.5 cm 2 . The former gives a choice of a best tool for the job in terms of modeling the penumbra vs the inner portion of the beam. The latter is beneficial in optimizing dosimetric accuracy of small MLC apertures found in highly modulated plans.
Our attempts to optimize the all-important leaf-end shape and offset parameters agree with various previous findings that the results based on simple static and dynamic fields depend on the measurement techniques and conditions 2,3,37 and ultimately highly modulated realistic plans are necessary to hone those values. 13,38,39 Of all variables in the RayStation model, we find the leaf tip width to be the least intuitive. It is an incremental attempt at modeling a rounded leaf end, in between the Eclipse flat end approach with a constant offset across the field 40 and Pinnacle ray-tracing through the rounded leaf end shaped and positioned almost as in the real world. 41 We failed to find a quantitative relationship between the optimal leaf tip width value and the physical leaf shape for various MLC models. On a practical level, our static field experiments indicated that this parameter had no appreciable effect on the Halcyon penumbra profiles, and it was left at 0. While the MLC Offset parameter was eventually optimized at +0.007 cm, the model-based Gain and Curvature values were so close to zero that the difference was attributed to experimental uncertainties. 14

| CONCLUSIONS
The self-contained Halcyon radiotherapy platform was successfully  APPENDIX Table 1 provides details of the measurement setup and instrumentation for the basic beam data collection (Section II.B.2).