Distributive quality assurance and delivery of stereotactic ablative radiotherapy treatments amongst beam matched linear accelerators: A feasibility study

Abstract Purpose Beam matching occurs on all linacs to some degree and when two are more are matched to each other, patients are able to be transferred between machines. Quality assurance of plans can also be performed “distributively” on any of the matched linacs. The degree to which machines are matched and how this translates to like delivery of plans has been the focus of a number of studies. This concept has not yet been explored for stereotactic techniques which require a higher degree of accuracy. This study proposes beam matching criteria which allows for the distributive delivery and quality assurance of stereotactic body radiotherapy (SBRT) plans. Method Two clinically relevant and complex volumetric modulated arc therapy (VMAT) SBRT spine and lung plans were chosen as benchmarking cases. These were delivered on nine previously beam matched linacs with quality assurance performed through ArcCheck and film exposure in the sagittal plane. Measured doses were compared to their treatment planning system predictions through gamma analysis at a range of criteria. Results Despite differences in beam match parameters and variations in small fields, all nine linacs produced accurate deliveries with a tight deviation in the population sample. Pass rates were well above suggested tolerances at the recommended gamma criterion. Film was able to detect dose errors to a greater degree than ArcCheck. Conclusion Distributive quality assurance and delivery of stereotactic ablative radiotherapy treatments amongst beam matched linacs is certainly feasible provided the linacs are matched to a strict protocol like that suggested in this study and regular quality assurance is performed on the matched fleet. Distributive quality assurance and delivery of SBRT provides the possibility of efficiency gains for physicists as well as treatment staff.

ear accelerators and performed beam match analysis by delivering standard plans on all the linacs. 8 The study showed that matching a new linac to a reference linac does not guarantee that it is matched to the established treatment planning model. Similarly, Gersgjevitsh et al. 9 demonstrated that a single TPS model could not be supported by their three Elekta linacs which were matched according to the vendor's beam matching criteria, calling for stricter criteria and an additional subset of dosimetric data to be matched to if a single model was to be used. Alternatively, a study by Swamy et al., like Sjöström et al., also considered "fine beam-matched" Varian linacs and found that from delivery of fifteen volumetric modulated arc therapy (VMAT) plans, the standard deviation in gamma (2%/2 mm) pass rates was 1.00%, demonstrating excellent beam matching in terms of VMAT delivery. 10 To date there have not been any studies published considering beam matching for stereotactic body radiotherapy (SBRT) and the tighter tolerances and objectives it might require, so the feasibility of distributive QA and treatment delivery of SBRT should thus be assessed and verified.
Stereotactic body radiotherapy describes extracranial treatment techniques which utilize a larger delivery of radiation dose than conventional radiotherapy and in fewer fractions resulting in a higher biological effective dose for the treatment site. 11 The hypofractionated nature of SBRT treatments provides a unique challenge for beam matched linacs, as any small differences in beam delivery will have a larger effect on the overall treatment. This work investigates the effect of current beam matching procedures on complex SBRT delivery by benchmarking nine nominally matched Elekta linacs against standard plans. This is of particular interest for SBRT spine treatments due to increased beam complexities involved. Of chief consideration is the involvement of small fields which have been shown to exhibit inconsistencies between otherwise dosimetrically matched linacs. 12

| MATERIALS AND METHOD
The tighter beam matching criteria used by the linacs in this study, which improves upon the vendor beam matching criteria and the criteria set out by Hrbacek et al., 5  Two clinical SBRT plans, one spine and one lung, were chosen as benchmarking cases for assessment of SBRT beam matching. Both were previously planned in Pinnacle 3® 9.10 (Koninklijke Philips N.V., Amsterdam, The Netherlands) and treated on an Elekta VersaHD ® linear accelerator (Elekta, Stockholm, Sweden). The cases were chosen as they were particularly complex compared to other SBRT plans of the same site according to their QA pass rates. 15 These plans were sent to the respective record and verify systems for treatment.
The spine case included four VMAT arcs with 120 control points per arc and a prescription of 30 Gy in three fractions to the planned target volume and a maximum dose to the spinal cord of 18 Gy. 16 A transverse slice of the original plan is shown in Fig. 1 In order to observe trends between nominally beam matched machines and assess suitability of moving SABR plans or QA between linacs, multiple distance-to-agreement (DTA) criteria were assessed and compared. Other characteristics were also quantified such as the dose gradient between target and central nervous system (CNS) and the CNS point dose difference in the case of spine as well as some machine settings like MLC offsets. Differences were quantified through examination of the standard deviations in the linac population sample.
Transverse slice of the original stereotactic body radiotherapy spine plan with the 95% isodose shown in red and the target and spinal cord in green and purple respectively.
Transverse slice of the original stereotactic body radiotherapy lung plan with the 95% isodose shown in red and the target green.
T A B L E 2 Matched beam parameters for linacs 1 to 9. Profiles assessed for a 30 × 30 cm 2 field at 10 cm depth in the in-plane (IP) and cross-plane (XP) directions. PDDs are taken for a 10 × 10 cm 2 field at either 100 or 90 cm SSD for beam quality matching. Both spine and lung cases passed ArcCheck and film QA on all machines tested according to routine clinical methods and recommended tolerances (>90% pass rate at 3%/2 mm). 19 The suggested gamma criteria of 3%/2 mm was used as the reference criterion for comparison between linacs for both film and ArcCheck. An example of a film dose comparison for the spine case assessed at 3%/2 mm is shown in Fig. 3 where the difference in dose gradient between delivered and planned (5.42%/mm) dose can be observed.
QA results are listed in Table 3 24 All target doses were within 5% of planned. [25][26][27] Aside from the CNS dose, the dose drop off between target and spinal cord is one of the most important aspects of SBRT spine QA. The variation in dose gradient was also found to be very 3. An example of comparison between measured (Set 1) and planned (set 2) sagittal plane dose profiles for the spine case delivered on linac 7. The twodimensional dose map comparison had a pass rate of 98.8% at 3%/2 mm T A B L E 3 Quality assurance results of both spine and lung cases.  | 103 plane while ArcCheck attempts to gain an appreciation of the total dose fluence. The two QA methods also differ in accuracy on a few levels. Film receives a higher maximum dose so is less sensitive to low dose differences (when using global comparison), suffers from processing and calibration uncertainties but has a higher resolution (scanned at 75 DPI) and can show where dose differences will likely occur anatomically. ArcCheck receives a lower maximum dose due to the fluence washout so is arguably more sensitive in regions with low dose fluence, can detect angular dependent fluence errors, has lower uncertainty in dose calibration but also has a lower resolution (detector spacing of 1 cm 29 ). Some difference in pass rates can thus be expected. Considering individual machines, linac 4 and 9 were initially outliers in results, failing in the case of spine. Despite performing ordinary VMAT QA to a satisfactory level of accuracy, dose profiles for the two SBRT cases initially appeared either shrunk or stretched. In these cases, the jaws and MLCs were recalibrated. The recalibration process, called the "Agility Workflow," was avoided in linacs 1 and 5 by instead applying sub-millimetre MLC offsets to their MLC calibrations. Both methods improved SBRT performance to produce passing results. These results suggest that additional criteria are required for SBRT beam matchingsimply to perform complex benchmarking cases such as those demonstrated in this study to observe any need for MLC/recalibration or offsets. This also indicates that, if a distributive QA and treatment delivery model is adopted, a weekly or monthly complex SBRT plan QA is required on all linacs to ensure that MLC calibration has not drifted. If the SBRT workload is high enough then each machine may perform enough QA weekly to avoid the need for this, but if the SBRT patient workload is light then this may stretch physics resourcesespecially if variations in bunker designs require a pretreatment run through of beams (e.g., to avoid couch collision with couch rotations).

Metric
Interestingly, Table 2 shows that the profile flatnesses are not tightly grouped about the baseline value, making use of their 1% wiggle room. This appears to have little effect on the QA results, perhaps for the sole reason that the test cases involved small, isocentric fields where profiles become increasingly indistinguishable, even between flattened and unflattened fields. Even for nonisocentric plans, the 3% dose difference criterion would not be sensitive enough to detect these profile differences. This then puts greater emphasis in the accuracy of the machine's MLC calibration and dose output.
In addition to the tests described here, other parameters need to be tighter than for regular treatments 30 such as the geometric accuracy of linac and couch as well as image guidance systems. These should also be re-evaluated before each treatment. The tests in this study were concerned with doses calculated in homogeneous water equivalent phantoms, so the dose calculation accuracy in heterogeneous anatomical phantoms should also be verified in a method such as that described in the beam matching process for the Elekta linacs in this study.
This study has shown that, for nine nominally matched Elekta linacs, SBRT spine and lung delivery was equivalent among linacs and well within accepted tolerances. This sample of linacs is thus a good candidate for moving toward the feasible goal of a distributive QA and treatment. The clinical impact of the calculated uncertainties is certainly within accepted limits. 26,27 For a department that wishes to beam match their linear accelerator fleet to a degree beyond VMAT that allows for distributive QA and delivery of SBRT, the following items are recommended: 1. Profiles and PDDs should be matched to criteria that is stricter than vendor guidelines. 9 For example, the linacs in this study were matched at commissioning with PDD 20,10 matching <0.5% and all points in the PDD within 0.5% of the reference curve. All open profile points within 80% of the field size were within 1% of reference 2. Comprehensive VMAT benchmarking should be performed for a variety of treatment sites, verifying a base level of beam matching traceable to the TPS model. 7 Level 3 testing should also be performed in anatomical phantoms.
3. kV-MV isocentre coincidence should be within specification for SBRT delivery for all linacs concerned. Ideally, this should be checked regularly. Image guidance, immobilization, and linac capabilities should, of course, be equivalent between machines.

4.
Complex SBRT plans can be used as benchmark cases with QA performed with film (and diode array) at a range of gamma criteria. If cases fail, then the jaw or MLCs may need to be recalibrated of offsets applied to the existing calibration. A diode array in not sufficient in itself for accurate benchmarking 5. QA results should all pass film QA with a small standard deviation, thus demonstrating a low uncertainty in accuracy for distributive QA and treatment delivery.
6. If the SBRT workload is light, then regular testing of each machine through a standard complex SBRT case should be performed to observe any drift in MLC calibration.

7.
Daily output and imaging checks should be performed during run-up or otherwise to capture any gross differences on the day of treatment.

| CONCLUSION
A study of complex SBRT lung and spine cases was conducted through delivery and quality assurance on nine beam matched linacs in order to assess the feasibility of a distributive approach to SBRT QA and delivery. The results of this study suggest that at recommended gamma criterion the delivery accuracy of the SBRT cases is well-matched between machines and well above the recommended gamma pass rates, with differences in pass rates at stricter criteria attributed to MLC discrepancies. While both film and ArcCheck are informative in the benchmarking process, film should always be used for QA of clinical plans. Distributive QA and delivery of SBRT is thus feasible but requires initial beam matching of machines at stricter criteria than recommended by manufacturers as well as an ongoing QA program that also includes assessments of other parameters like image guidance.

ACKNOWLEDGMENTS
Thanks go to Jason Arts and Simon Goodall for measurement of some cases.

CONF LICTS OF INTEREST
None.