Validation of the preconfigured Varian Ethos Acuros XB Beam Model for treatment planning dose calculations: A dosimetric study

Abstract Varian (Palo Alto, California, United States) recently released an online adaptation treatment platform, Ethos, which has introduced a new Dose Preview and Automated Plan Generation module despite sharing identical beam data with the existing Halcyon linac. The module incorporates a preconfigured beam model and the Acuros XB algorithm (Ethos AXB model) to generate final dose calculations from an initial fluence optimization. In this study, we comprehensively validated the accuracy of the Ethos AXB model by comparing it against the Halcyon AXB model, the Halcyon Anisotropic Analytical Algorithm (AAA) model, and measurements acquired on an Ethos linac. Results indicated that the Ethos AXB model demonstrated a comparable if not superior dosimetric accuracy to the Halcyon AXB model in basic and complex calculations, and at the same time its dosimetric accuracy in modulated and heterogeneous plans was better than that of the Halcyon AAA model. Despite the fact that the same algorithm was utilized, the Ethos AXB model and the Halcyon AXB model still exhibited variations across a range of tests, although these variations were predominantly insignificant in the clinical environment. The accuracy of the Ethos AXB model has been successfully verified in this study and is considered appropriate for the current clinical scope. On the basis of this study, clinical physicists can perform a data validation instead of a full data commissioning when implementing the Ethos system, thereby adopting a more efficient approach for Ethos installation.

While different algorithms are used at different stages of plan generation, the final dose calculation in Ethos is performed by an implementation of the AXB algorithm previously introduced in another treatment planning system (TPS), Varian Eclipse. Although the accuracy of the AXB algorithm in the Eclipse TPS has been verified by multiple studies, [3][4][5][6][7][8] it has been implemented into a dedicated environment (the Ethos TPS) using dedicated computing hardware.
Previous investigations between AXB calculations with central processing unit (CPU) and graphical processing units (GPU) demonstrate the accuracy of AXB with GPU calculations, 9 however, no such testing exists in the literature for the Ethos platform.

Varian has introduced a preconfigured beam model for Ethos and
Halcyon, which is created based on the 6 MV FFF golden beam data with the Ethos TPS, and its implementation in the Eclipse environment has been extensively verified. [3][4][5][6][7][8] Calculations were also compared to the Halcyon AAA model, because its accuracy was not only verified by authors during the initial commissioning but has also been validated in several other studies. [12][13][14] A similar approach was adopted by a previous study during the validation of the Eclipse AXB algorithm. 15 All calculation results were also compared to measurements acquired on an Ethos linac.

| MATERIALS AND METHODS
The AXB algorithm requires the macroscopic atomic cross sections of the components of the dose calculation material. 16 Ethos utilizes a dual-multi-leaf-collimator (MLC) design that has a maximum square field size of 28 × 28 cm 2 . Therefore, in this study, all verification tests were limited to field sizes smaller than or equal to 28 × 28 cm 2 . All fields measured in this study were defined by MLCs as the machine does not have jaws. All comparisons were conducted using the 6 MV flattening filter-free (FFF) energy, the only available energy on both Ethos and Halcyon linacs.

2.A | Basic tests
Treatment fields for basic tests were created in Eclipse in a 40 × 40 × 40 cm 3 virtual water tank. Each field was computed using the Halcyon AXB and AAA models, respectively. Subsequently, the plan and its associated structures and datasets were exported to the Ethos TPS to recalculate using the Ethos AXB model. The calculation results were then compared to measurements.

2.B.1 | Chair tests
A set of five chair test plans were optimized in a homogenous water volume. Modulation characteristics of each plan were controlled by implementing incremental X and Y smoothing parameters for IMRT optimization. Smoothing parameters ranged from X = 10, Y = 20 to X = 60, Y = 70, respectively. Each plan was optimized with the same dose objectives for the chair-shaped target and surrounding avoid structures. The increase of modulation in each of the respective plans was evaluated by total plan MU per target dose (MU/Gy).
The plans were subsequently delivered to homogeneous solid water slabs, in which a PTW Semiflex 3D chamber was inserted. The measured chamber reading was then compared to the calculated point dose at the corresponding position.

2.B.2 | Clinical plans in a homogeneous phantom
A set of clinical plans of different anatomical sites, including abdomen, brain, bilateral head and neck, unilateral head and neck, and prostate, were retrospectively selected and recomputed on the Arc-Check phantom in both TPSs. In both systems, the density of the ArcCheck was overridden to "Poly(methyl methacrylate) (PMMA)" (1.19 g/cm 3 ). Once computed, plans were delivered to ArcCheck, followed by comparing the measured and the calculated dose distributions in the SunNuclear SNC Patient software. In addition, a comparison was also performed directly between the three models.
The gamma criteria used for analysis were 3% and 2 mm. 22 During measurements, a PTW Semiflex 3D chamber was inserted at the isocenter of the ArcCheck to measure point doses.

2.C | Advanced tests to investigate the management of tissue heterogeneity
In this section, two clinical plans, a VMAT and an IMRT plan, were computed in both TPSs on a CIRS thorax phantom, which consisted of tissue, lung, and bone materials. A Standard Imaging (Middleton, Wisconsin, United States) Exradin A1SL chamber was used to take measurements at the center of the phantom and compare to the calculation results. In both plans, the beam arrangement meant that most of the dose was delivered through significant heterogeneity, as shown in Fig. 1.
Phantom images used in planning were scanned with the chamber insert, and therefore the position of the chamber was air in these images. AXB explicitly models the physical interactions of radiation and matter, whereas AAA use pre-calculated MC kernels scaled according to local density variations. 16 In addition, the Ethos TPS can only report dose to medium (D m ), which in this case would report dose in air. However, when appropriate correction factors are applied, chambers measure dose to water (D w ). In order to compare calculated and measured doses, in both the Ethos TPS and the Eclipse TPS, the chamber hole was overridden to the material of water (1.0 g/cm 3 ). In addition, as Ethos does not allow the report of per beam dose, for both the IMRT and the VMAT plans, integral plan doses were instead reported and compared.
In addition to point doses, a vertical line dose profile (going through the target and the bone) and a horizontal line dose profile (going through the target and both lungs) were plotted and compared between the Ethos AXB model and the Halcyon models.

2.D | Statistical analysis and uncertainty
Where statistical analysis was required, a one-tailed t-test was performed provided the number of samples was large enough. P < 0.05 was considered statistically significant.
For chamber measurements that were acquired together with an electrometer, the uncertainty was calculated using the formula below: where ɛ a and ɛ b are the estimated measurement uncertainty of each component (chamber and electrometer) in the measurement system.
In this paper, three chambers were used during measurements, which were: PTW microdiamond, Semiflex 3D, and Exradin A1SL.
The individual long-term stability of the PTW microdiamond and Semiflex 3D detector was 0.25% 23 and 0.30% 24 according to their technical specifications. The long-term stability of the Exradin A1SL chamber was not specified by the vendor but was estimated to be 0.5% based on authors' experience with the chamber. For all measurements, a PTW Unidos Webline electrometer was used, the uncertainty of which was 0.5% .25 Therefore, using the above formula, the combined uncertainty of the PTW microdiamond, the Semiflex 3D, and the Exradin A1SL measurement system was calculated to be 0.6%, 0.6%, and 0.7%, respectively. As imaging was used in all measurement setups, setup uncertainty was considered negligible and therefore excluded in the uncertainty analysis.     3.C | Advanced tests to investigate the management of tissue heterogeneity   showed an agreement similar to that of the Ethos AXB model, which was within AE1.5% of measurements beyond the buildup region for field sizes > 1 × 1 cm 2 , and within AE2.0% beyond the buildup region for the field size of 1 x 1 cm 2 . However, for most field sizes, the calculated PDD was slightly lower than the measured PDD. This difference between the Ethos AXB model and the Halcyon AAA model is more obvious when plotted in the same graph (Fig. 16), especially towards the tail of the curve.

3.B.2 | Clinical plans in a homogeneous phantom
Since several studies have suggested that the AAA algorithm and the AXB algorithm demonstrate good agreement in water, 4 Profile comparisons indicated that for the Halcyon AXB and AAA models, the agreement between the calculated and the measured profiles was between AE1.5% in the umbra region for all field sizes.
On the contrary, for the Ethos AXB model, although calculated profiles of larger field sizes (10 × 10 and 28 × 28 cm 2 ) showed similar agreement in the umbra region, this was not the case for the 4 × 4 cm 2 field, where larger differences were observed. In addition, the shape of the calculated profile was slightly wider than that of the measured one, as shown in Fig. 17.   29 In addition, pairwise comparisons indicated that differences between these three models were not statistically significant (P > 0.05). Interestingly, for all three models, larger deviations were observed at the 2 × 2 cm 2 and the 3 × 28 cm 2 fields, which could be an intrinsic modelling issue of preconfigured Halcyon/Ethos models.

4.B | Advanced tests for specific planning or treatment conditions
To test the performances of both models in calculating IMRT/VMAT in plans in a homogeneous material, the chair test and a set of clinical plans were utilized, with measurements acquired using both a single-point method and a 3D method. 22 In the chair test, the Hal- heterogeneous tissues or tissue interfaces, it was most likely due to differences in the algorithm. It can also be seen that the Halcyon AAA model tends to predict higher doses in the lung/tissue interface, which is consistent with previous literature. 31,34 Although the tissue heterogeneity test in this paper was rela-

CONFLI CT OF INTEREST
The authors declare that there is no duality of interest that they should disclose.