Identification of a potential source of error for 6FFF beams delivered on an AgilityTM multileaf collimator

Abstract Purpose The performance of the AgilityTM multileaf collimator was investigated with a focus on dynamic, small fields for flattening filter free (FFF) beams. Methods In this study we have developed a simple tool to test the robustness of the control mechanisms during dynamic beam delivery for Elekta’s VersaHD linear accelerator with Integrity 4.0.4 control software. We have programed the planning system to calculate dose for delivery of sweeping gaps. These sweeping gaps have a constant speed, constant size, and are delivered at a constant dose rate. Therefore they specifically identify delivery problems in dynamic mode. Results The Elekta AgilityTM control mechanism fails to maintain accurate delivery for small, dynamic sweeping gaps. For small gap sizes, the AgilityTM control mechanism delivers a field that is more than four times the size of the planned field width without generating an interlock. This has dosimetric implications: The discrepancy between calculated and measured doses increases with decreasing gap size and exceeds 10% and 60% at isocenter for a 3.5 mm and 1 mm gap size, respectively. Conclusion A deficiency of the AgilityTM control system was identified in this study. This deficiency is a potential source of error for volumetric modulated arc therapy fields and could therefore contribute to relatively high failure rates in quality assurance measurements, especially for FFF beams.

TPS as the high dose rate accentuates the impact of small delivery errors.
In this study we focus on delivery anomalies of the Agility TM MLC, which could contribute significantly to discrepancies between calculated and measured doses for regular clinical plans.
In order to identify relatively high failure rates on routine quality assurance (QA) measurements, specific MLC test patterns were developed in order to isolate potential root causes. The evaluation of delivered MLC fields has triggered the investigation presented in this manuscript, which focuses on the delivery accuracy of the Agility TM MLC for small aperture in dynamic mode. The same dynamic MLC sequences were also delivered on Varian's TrueBeam TM for comparison.

| MATERIALS AND METHODS
The 6FFF beam modality on Elekta's VersaHD TM linear accelerator (linac) with Integrity 4.0.4 control software, was fully commissioned and the corresponding beam model validated in Elekta's Monaco® TPS, version 5.11.02. 8 Beam tuning, MLC calibration, and MLC beam modeling in Monaco® were performed according to Elekta's recommended procedure, taking into account most recent findings. 9 Performance of these procedures was well within required specifications. MLC parameters in Monaco® were fine-tuned to achieve optimal agreement for point dose measurement and dose distribution for a range of VMAT and dynamic conformal arc therapy (DCAT) plans. 10 However, clinical VMAT plans still showed unacceptably low pass rates on the VersaHD TM linac, especially for highly modulated fields and small targets. SunNuclear's ArcCheck TM was used for these measurements.

2.A | Sweeping gap test fields
We have developed test fields, which consist of dynamic MLC (dMLC) sequences. Each sequence represents a gap of constant size, sweeping across the field at constant speed and dose rate. The gap size is different for each field (20 mm, 10 mm, 5 mm, 3.5 mm, 2 mm, and 1 mm). Similar tests have been used, 11 including their application for FFF beams, 12 but, to the best of the author's knowledge, have never been measured on a VersaHD TM with Agility TM MLC.
These fields specifically identify delivery problems in dynamic mode and potential weaknesses of certain parameters in the TPS, such as the leaf offset, leaf tip transmission, interleaf transmission, and MLC scatter. 13 All fields were delivered with 3000MU, except for the 20 mm gap (2000MU). The dose calculations were performed with Monaco's XVMC Monte Carlo algorithm using a 1.5 mm grid size and a variance of 0.5% per control point.
These test fields were generated in Monaco 5.11.02 and calculated on a PTW RW3 phantom. All fields were delivered in clinical mode with a nominal dose rate of 1400 MU/min using the Mosaiq® Radiation Oncology patient management system. Measurements were performed using IBA's MatriXX® Evolution detector and compared with calculated dose distributions.
The service graphing tool was used to evaluate the so-called "desired" and "actual" leaf positions during beam delivery. The data points were acquired with a sampling rate of 4 Hz.

2.B | Clinical plans
Further to these described test fields a range of VMAT and DCAT plans were created by optimizing patient plans in Monaco® using its built-in sequencer. These plans created by Monaco® were analyzed with in-house software in order to quantify the contribution of small leaf apertures (≤3.5 mm). All leaf apertures not covered by the diaphragms were quantified for all exposed leaf pairs and segments and weighed with the corresponding MU per segment, following an approach described by Feng et al recently. 7 The MU-weighed ratio of small apertures (≤3.5 mm) to the total number of apertures was then determined for each plan.
This selection of DCAT and VMAT plans was also measured, using the locally established QA procedure: Measured with SunNuclear's ArcCheck TM in absolute mode and evaluating the 2%/2 mm as well as the 3%/3 mm gamma criteria. The option to "Apply Measurement Uncertainty" in the corresponding software was selected.

3.A | Sweeping gap test fields
It was found that the dose calculation for gap sizes of 20 mm, 10 mm, and 5 mm is in agreement with measurements. The relative dose difference at central axis was −2.1%, −1.0%, and −0.8% for the 20 mm, 10 mm, and 5 mm gap, respectively (negative value indicates lower measured dose). However, for gap sizes 3.5 mm and below significant errors were identified, as shown in  +1.5 mm (maximum 1.9 mm) for X1 and X2, respectively. This introduces a total error of 3mm for the 1mm gap size, therefore quadrupling the field width.
In comparison, the TrueBeam TM delivered all these fields according to plan and dosimetric agreement was well within specifications for all gap sizes (see Fig. 4).

3.B | Clinical plans
As the Agility TM seemed unable to maintain accurate leaf positioning for aperture widths of less than 3.5 mm, other Monaco® plans were analyzed in terms of aperture widths and passing rates.

| DISCUSSION
There are three major questions that should be asked when considering these data:  1. Why did the VersaHD TM delivery system not generate an interlock when the described delivery error occurred?
2. Do such small gaps (3.5 mm or smaller) occur in clinical plans?

Does this impact pass rates for clinical plans?
The following discussion will consider these three questions and recommend further work that could be undertaken.

4.A | No interlock
The dynamic tolerance for leaf control is set to 1.0 mm in Integrity According to Rangel and Dunscombe systematic leaf positioning errors must be limited to 0.3mm in order to achieve delivered dose accuracy within 2 Gy for organs at risk and within 2% for the equivalent uniform dose to the target. 17 As even EPID imaging verification tools can detect systematic MLC errors as small as 3 mm 18 the expectation for the MLC controller would be to be far more sensitive to leaf positioning errors than this. However, the data presented in our study seem to suggest that the Agility TM used in dynamic mode might not generate an interlock for errors as large as 1.5 mm.

Comprehensive analysis by Kerns et al. has determined a RMS
leaf position accuracy of 0.32 mm for dynamic treatment fields. 19 Wang et al. have found the leaf positioning accuracy on an Elekta Synergy linac largely to be within 0.5 mm. 20 Both of these studies, however, have not focused on small apertures in dynamic mode.

4.B | Small gaps in clinical plans
Tatsumi et al. have analyzed the MU-weighted segmental average of the mean leaf gap width and correlated this with average pass rates. 21 Our study suggests that the minimum gap width, as prescribed by the TPS, may be a sensitive metric to predict pass rates.
We have determined the MU-weighted ratio of gaps that are 3.5 mm or less to the total amount of MU-weighted apertures. For representative clinical VMAT plans we found this ratio to be anywhere between 0.05% and 10%. For more complex cases, like a single-isocenter multiple brain metastasis treatment, this ratio can exceed 24% (see Fig. 5). Depending on the location of these 3.5mm gaps, they could introduce a relatively large error in delivered dose.

4.C | Impact on pass rates
Multiple studies have demonstrated the impact of MLC positioning errors on the accuracy of dose delivery. [22][23][24][25] Even though it was shown that VMAT is less susceptible to delivery errors than dMLC treatments, 26  Dunscombe proposed a 0.3 mm limit. 17 This study demonstrates that the Agility TM control system does not comply with these suggested limits. The same dynamic MLC sequence delivered on the TrueBeam linear accelerator shows good agreement. This indicates that the problem does not lie with the MLC parameters used by the TPS, but rather with the Agility TM control system. This observation is also confirmed by a log file analysis.
The linear regression shown in Fig. 5 suggests that there could be some correlation between passing rates and ratio of small segments in any given plan.

4.D | Future directions
There are multiple avenues that can be investigated. First, a multi-institutional evaluation of planned vs actual leaf positions during delivery, as has been published already for Varian machines. 19 Second, to introduce systematic and random MLC positional or speed errors in representative patient plans to determine the Agility TM control system's threshold for error detection. Subsequently, the impact on dose delivery and dose volume histogram parameters for representative patient plans can be quantified.
While all this work will certainly be interesting to undertake, we believe that the current state of this project warrants sharing among the wider Medical Physics community, as we have concerns about the performance of the Agility TM control system.
Unless leaf positioning accuracy can be guaranteed during beam delivery, the possibility of mistreatments on the Agility TM is very real, especially when a TPS does not take the identified Agility TM limitations into account. Each center must therefore set appropriate action levels in their quality assurance program to identify potential delivery problems.

| CONCLUSION
A deficiency of the Agility TM control system was identified in this study, showing that beam delivery continues without interlock despite exceeding performance specifications. The resulting maximum dose difference at the central axis was observed to be up to 10.2% and 61.3% for a small field of 1 mm and 3.5 mm width, respectively. These findings are of concern and warrant imposing additional quality assurance measures to ensure patient treatment is as safe as possible.

CONF LICT OF I NTEREST
No conflict of interest.