A Publicly Available Dataset of Out-of-Field Dose Profiles of a 6 MV Linear Accelerator

An increase in radiotherapy-induced secondary malignancies has led to recent developments in analytical modelling of out-of-field dose. These models must be validated against measurements, but currently available datasets are outdated or limited in scope. This study aimed to address these shortcomings by producing a large dataset of out-of-field dose profiles measured with modern equipment. A novel method was developed with the intention of allowing physicists in all clinics to perform these measurements themselves using commonly available dosimetry equipment. A standard 3D scanning water tank was used to collect 36 extended profiles. Each profile was measured in two sections, with the inner section measured with the beam directly incident on the tank, and the outer section with the beam incident on a water-equivalent phantom abutted next to the tank. The two sections were then stitched using a novel feature-matching approach. The profiles were compared against linac commissioning data and manually inspected for discontinuities in the overlap region. The dataset is presented as a publicly accessible comma separated variable file containing off-axis ratios at a range of off-axis distances. This dataset may be applied to the development and validation of analytical models of out-of-field dose. Additionally, it may be used to inform dose estimates to radiosensitive implants and anatomy. Physicists are encouraged to perform these out-of-field measurements in their own clinics and share their results with the community.


Introduction
Improvements in diagnostic and treatment technologies have resulted in overall cancer survival rates increasing in recent times [1]. However, this increase in radiation therapy survivorship has coincided with a commensurate increase in the rate of radiation induced secondary malignancies later in life [2]. In response, there has been increasing effort to better understand out-of-field doses delivered during radiation therapy treatments [3]. Beyond secondary cancer induction, accurate out-of-field dose estimates are also necessary for assessing doses to radiosensitive implants such as cardiac devices [4,5] and radiosensitive anatomy as in the case of pregnant patients [6,7]. There is also a need for increasingly accurate out-of-field dose estimates to inform epidemiological studies [8].
Methods of estimating out-of-field doses include consulting simple reference data in the literature [6,[9][10][11], Monte Carlo simulations [12][13][14], and increasingly refined analytical models [15][16][17][18]. These models may need to consider the physical geometry of clinical linacs, including jaws, primary collimator, MLC leaves and carriage, and the arrangement of additional head shielding. This is further complicated by the relative positions of the linac treatment head and patient, such as during non-coplanar cranial radiotherapy [19]. Modelling and verifying the effects of these factors would require the measurement of a large number of out-of-field profiles with many collimator orientations. As out-offield dose modelling matures, it is also reasonable to expect that computational models will augment traditional dose calculation algorithms in commercial treatment planning systems. Commissioning these models may require the measurement of profiles much farther outside the field than physicists are currently accustomed.
Available datasets in the literature generally contain older model linacs and MLCs that are rarely seen today [6,9] and present coarse resolution dose profiles in a limited set of collimator orientations. Physicists wishing to use these profiles are forced to interpolate values from the printed figures by hand. This is a sub-optimal approach, and a contemporary solution using modern dosimetry equipment should achieve much greater spatial resolution while also presenting the data digitally in a manner that is computationally digestible.
In this article we establish a method of measuring high-quality out-of-field dose profiles using typically available clinical physics equipment. This method is then followed to produce a comprehensive and publicly available dataset of out-of-field dose profiles of our clinical linac.

Equipment
The linac under investigation in this study was a Varian Clinac iX (Varian Medical Systems, Palo Alto, USA) with a Millennium 120 MLC. All measurements were performed using the 6 MV photon energy with 600 MU/min dose rate and flattening filter. A PTW Semiflex 0.3 cm 3 31013 (PTW-Freiburg, Freiburg, Germany) ionisation chamber was used for all profile scans. The dimensions of this chamber afforded lower volume averaging compared to larger thimble chambers, while still maintaining the sensitivity necessary for far outof-field measurements. The chamber was affixed to a PTW BEAMSCAN water tank using the TRUFIX chamber positioning system and steered with version 4.3 of the control software. The BEAMSCAN water tank had a usable scanning range of 500 mm (horiz.) × 500 mm (horiz.) × 415 mm (vert.), and 15 mm thick PMMA walls. A PTW Semiflex 3D 0.07 cm 3 31021 ionisation chamber was used as a reference to correct for fluctuations in the beam output during each scan. The chamber voltages were set to 400 V as per manufacturer recommendations, and the integrated water tank electrometer was set to low range. The control software was programmed to measure profiles in continuous scanning mode with a chamber speed of 5 mm/s and data points reported every 2 mm. A 30 × 30 × 40 cm 3 stack of Virtual Water (Standard Imaging, Middleton, USA) was used as additional scattering material.

Measurement Setup
Each profile was measured as a piece-wise combination of two scans to construct a much longer profile measurement than a typical water tank would allow. The first section acquired the in-field and near out-of-field region of each profile, while the second section captured the far out-of-field region. An overlap area of approximately 15 cm was included in each pair of profile sections.
The first measurement geometry can be seen in Figure 1a. To begin, the tank was positioned such that the long axis of the chamber was orthogonal to the scanning direction, with a source to surface distance of 90 cm. The tank was then translated such that the central axis of the beam was 15 cm from the maximum chamber travel position on one side, and 35 cm from the maximum chamber travel position on the opposite side. This allowed the inner section of each profile to be collected out to 35 cm from the central axis while still allowing adequate scatter around the field.
The measurement geometry was then altered to capture the far out-of-field section of each profile, as seen in Figure 1b. The tank was translated 31.5 cm in the direction of the profile, and a stack of Virtual Water was positioned in the field at 90 cm source to surface distance to re-establish full scattering conditions. Care was taken to ensure that the Virtual Water was abutted firmly against the tank wall with minimal air gaps.

Post-processing
The two sections of each profile were stitched together to create a whole.
This was non-trivial, as the overlap regions did not necessarily coincide due to small air gaps between the Virtual Water and the tank wall, the non-waterequivalence of the tank wall, and positioning errors in the tank. To overcome this issue, each overlap region was searched for a prominent feature common to both sections of the profile. The far out-of-field section was then progressively shifted in 1 mm increments and re-scaled until the identified feature matched the inner profile section as well as possible. The two sections were then combined into one full profile, with an average being taken in the overlap region. A visual example of this process can be seen in Figure 2. No additional smoothing or filtering was applied. The full profiles were validated by comparisons with commissioning data and visual inspections for any discontinuities.

Dataset Accessibility and Usage Notes
The

Dataset Validation
In order to establish that the irradiation conditions and measurement setup were representative of normal practice, the inner component of the 10 × 10 cm 2 X and Y profiles were compared against the same profiles gathered during commissioning of the linac. A gamma comparison of the in-field sections, bounded by the 50% isodose lines, showed 100% agreement with criteria of 1% dose difference (local normalisation) and 1 mm distance-to-agreement. This gave good confidence that the measurement equipment was set up correctly, that the linac was behaving nominally, and that ultimately the data gathered in the session was representative of clinical practice.
During the post-processing step, the outer section of each profile was shifted to coincide with the inner section of the profile. Given that the equipment was not moved during the measurement of all outer profile sections within a given orientation (radial or transverse), it follows that the ideal shift should be the same across all profiles. This was indeed found to be the case with the largest difference in ideal shifts being 1 mm. After combining the profile halves and taking the average in the overlap region, all full profiles were then manually inspected to ensure that there were no obvious discontinuities in the overlap region. None were found.

An Exploration of Dose Outside the Treatment Field
This investigation produced a large dataset with many curious features. In this section we compare and contrast a number of profiles that illuminate the underlying geometry and radiation interactions. However, these are just a representative selection of the profiles, and we encourage interested readers to further examine the dataset for themselves. Feature A denotes a sharp increase in relative dose at approximately 20 cm off axis, due to the finite lateral extent of the MLC. This feature has been observed in an earlier study [10]. This is immediately followed by a decrease in relative dose due to the primary collimator, which can also be seen on out-of-field. By this point most of the phantom scatter has been attenuated, leaving only leakage through the primary collimator and head shielding.
The 10 × 10 cm 2 MLC defined profiles were measured at depths of 10, 15, 20, and 25 cm to investigate the variation of off-axis ratio with depth. Figure   5 displays the X profiles normalised to the central axis dose of the profile at 10 cm depth. Remarkably, the out-of-field dose differs only marginally between depths. This indicates that out-of-field doses are largely independent of depth, at least at the distances and depths measured in this study. This finding is in line with earlier studies [6], but may not hold closer to the surface due to electrons scattered from the treatment head [12].  All profiles were measured in both the radial and transverse directions. Differences were only noted very far out-of-field. In all cases, the general shape of the relative dose out-of-field was similar between radial and transverse measurements, however, the transverse relative doses were of greater magnitude. This was best exemplified in the 5 × 5 cm 2 MLC defined configuration, shown in Figure 8 (feature A). This effect may be explained by an asymmetry in the linac head shielding.

Limitations
Many aspects of the above discussion involve observing features in out-of-field dose profiles and relating those features to the internal geometry of the linac head. There are many assumed relationships that cannot be stringently verified without Monte Carlo simulations that include a high-fidelity reproduction of the linac head, including the dimensions and locations of shielding blocks.
Linac vendors would need to be willing to share detailed 3D models to facilitate this as the phase spaces and simplified geometries commonly shared would not be suitable for modelling complex interactions in the head shielding. Such an arrangement has been possible in the past [12][13][14], and we encourage all vendors to be open to sharing this information into the future.
The measurements reported in this study were performed in a radiotherapy treatment clinic with a single model of linac, and so the extent to which this dataset can be extrapolated to other linac models is not obvious. It is reasonable to expect that the near out-of-field results may align with other models with similar tertiary collimation systems, such as the Varian Truebeam with Millennium 120 MLC. However, far out-of-field the results are more likely to depend on the exact head shielding arrangements, and so the similarity between systems in unclear.
Linacs with markedly different collimation systems, such as those produced by Elekta (Stockholm, Sweden), may have substantially different out-of-field features. Ultimately, we recommend that clinics gather their own dataset using the techniques presented in this study. We note however that our method would need substantial modification for use with cylindrical water tanks such as the 3D SCANNER (Sun Nuclear Corporation, Melbourne, USA), but should be directly transferable to other square tanks.

Recommendations
We recommend that this dataset is used to further investigate and understand features of out-of-field dose distributions, particularly by assisting in the creation and validation of computational models. Furthermore, we encourage readers to perform their own measurements and compare and contrast them with this dataset.
When planning the tank shift, we recommend that the overlap region of the inner and outer profile sections be centred around 20 cm from the central axis. This region contains several strong features, such as the beginning of the primary collimator, which allows the profiles to be stitched with confidence.
For clinics wishing to save time by collecting a more limited dataset for radiation protection calculations, we recommend collecting a single profile for a variety of field sizes. We believe the ideal profile is under the Y jaw, with the MLC fully retracted, in the radial orientation. This specific profile is compared against all others in Figure 9. From the figure it can seen that this profile represents an upper bound of the out-of-field dose for much of the range covered in this study, and can therefore be used as a conservative estimate of the dose across this range. Being in the radial direction, it is also representative of the majority of radiotherapy treatments.
We also make the suggestion that this same profile may be a good candidate for aiding in the development of out-of-field dose calculation models. It is a good basic test case as it is the only profile uncoupled from extra MLC shielding. MLC shielding may then be added as a second order effect. The jaw defined Y profiles for the three field sizes in this study (solid lines). The shaded areas correspond to the complete range of profiles measured for each field size.

Conclusion
This article has presented a comprehensive dataset of dose profiles outside of the treatment field. These profiles were collected in a non-academic radiotherapy treatment centre, with standard equipment, using a technique that should be repeatable elsewhere. Ultimately, we encourage other clinics to collect their own datasets and make them freely available to the community. Access to highquality collections of out-of-field profiles for a range of contemporary linacs would be extremely useful for risk assessments, radiation protection studies, and the development and commissioning of out-of-field dose calculation models.