First experimental demonstration of VMAT combined with MLC tracking for single and multi fraction lung SBRT on an MR-linac

Background and purpose: VMAT is not currently available on MR-linacs but could maximize plan conformality. To mitigate respiration without compromising delivery efﬁciency, MRI-guided MLC tumour tracking was recently developed for the 1.5 T Unity MR-linac (Elekta AB, Stockholm, Sweden) in combi- nation with IMRT. Here, we provide a ﬁrst experimental demonstration of VMAT + MLC tracking for several lung SBRT indications. Materials and methods: We created central patient and phantom VMAT plans (8 (cid:1) 7.5 Gy, 2 arcs) and we created peripheral phantom plans (3 (cid:1) 18 & 1 (cid:1) 34 Gy, 4 arcs). A motion phantom mimicked subject- recorded respiratory motion ( A =11 mm, f =0.33 Hz, drift =0.3 mm/min). This was monitored using 2D-cine MRI at 4 Hz to continuously realign the beam with the target. VMAT + MLC tracking performance was evaluated using 2D ﬁlm dosimetry and a novel motion-encoded and time-resolved pseudo-3D dosimetry approach. Results: We found an MLC leaf and jaw end-to-end latency of 328.05( (cid:3) 3.78) ms and 317.33( (cid:3) 4.64) ms, which was mitigated by a predictor. The VMAT plans required maximum MLC speeds of 12.1 cm/s and MLC tracking superimposed an additional 1.48 cm/s. A local 2%/1 mm gamma analysis with a static measurement as reference, revealed pass-rates of 28–46% without MLC tracking and 88–100% with MLC tracking for the 2D ﬁlm analysis. Similarly, the pseudo-3D gamma passing-rates increased from 22– 77% to 92–100%. The dose area histograms showed that MLC tracking increased the GTV D 98 % by 5– 20% and the PTV D 95 % by 7–24%, giving similar target coverage as their respective static reference. Conclusion: MRI-guided VMAT + MLC tracking is technically feasible on the MR-linac and results in highly conformal dose distribution. (cid:1) 2022

A viable alternative for surgery to treat early stage non-small-cell lung cancer is stereotactic body radiation therapy (SBRT) [1,2]. The dose distributions need to be highly conformal around the target to ensure coverage, while sparing adjacent organs at risk (OARs) [3]. Although Intensity Modulated Radiation Therapy (IMRT) produces conformal dose distributions, Volumetric Modulated Arc Therapy (VMAT) can in some cases spare even more tissue with equal target coverage [4]. On C-arm linacs, VMAT typically reduces delivery times when combined with high-dose rate flattening-filter free beams [5]. These benefits make VMAT a standard delivery technique for lung SBRT. However, both the 0.35 T MRIdian (ViewRay Inc., Oakwood, USA) and the 1.5 T Unity MRlinac (Elekta AB, Stockholm, SWE) exclusively support fixedbeam IMRT delivery in clinical mode. To show that the MR-linac is capable of delivering VMAT plans, a recent study explored VMAT deliveries on Unity for prostate SBRT [6].
The dose conformity of IMRT and VMAT increases the need for respiratory motion management. Current options to achieve this include: internal target volume (ITV; Unity) [7], or gating (MRIdian) [8]. The ITV provides GTV coverage by covering the full tumour motion, which for lung SBRT is typically 1-3 cm in cranial-caudal (CC) direction [9][10][11]. ITVs are relatively large and might overlap with OARs near central tumours, increasing toxicity risks [12,2]. Contrastingly, gating irradiates the tumour only during parts of the breathing cycle, requiring smaller treatment margins [13,8]. However, gating also drastically extends beam-on times with only 50-60% duty cycle [14,15], which is especially concerning for single fraction treatments [14]. Gating efficiency can be improved by breath-hold techniques, but this requires patient compliance [16]. An alternative is multi-leaf collimator (MLC) tracking, which realigns the treatment beam with the most recently observed (or predicted) tumour position [17,18] tracking uses similar treatment margins as gating, while maintaining a 100% duty cycle. MLC tracking is well established in the context of C-arm linacs [19][20][21]. Previous experiments demonstrated MRI-guided MLC tracking on the MR-linac either for artificial [22,17] and subject-recorded [23] respiratory CCmotion for a single gantry angle with a conformal treatment field, or for IMRT treatments with artificial CC-motion [24]. To our knowledge, MLC tracking has not been demonstrated before on the MR-linac in combination with VMAT.
Combining VMAT with MRI-guided MLC tracking is complex, because the leaf and jaw motion of both dynamic modes is superimposed. Previous studies investigated the technical feasibility of VMAT + MLC tracking for C-arm linacs [25][26][27] or demonstrated its feasibility in clinical trials [28]. Besides the ability to directly VMAT combined with MLC tracking on an MR-linac monitor the target motion with MRI, the unique engineering setup of the MR-linac differs from any C-arm linac. Differences include: very fast gantry rotation, very fast MLC speed (>12.1 cm/s), fixed collimator angle (270 ), low dose rate (425 MU/min), and the ability to continuously rotate; which necessitate a detailed study of VMAT + MLC tracking on the MR-linac.
In this study, we provide a first experimental demonstration of VMAT combined with MLC tracking for a range of lung SBRT fractionation schemes. The VMAT + MLC tracking performance both for phantom and patient treatment plans was evaluated using 2D film dosimetry setups and a novel motion-encoded and time-resolved pseudo-3D dosimetry approach.

Materials and methods
All experiments were performed on an Elekta Unity MR-linac, featuring a 1.5 T MR scanner and a 7 MV linac. The radiation beam is shaped in CC direction using a 160-leaf MLC consisting of 7.125 mm wide leaves with fixed collimator angle, while dynamic jaws limit the radiation field perpendicular to the leaf travel direction.
In our experimental setup we used the Quasar MRI 4D phantom (Modus Medical Devices Inc., London ON) and the Delta 4 Phantom+ (ScandiDos, Sweden) for dosimetry.

VMAT + MLC tracking
This work used interfaces developed for research to combine VMAT [6] with MRI-guided MLC tracking [24]. The interfaces contain an optimized proportional-integral-derivative (PID) control loop of the MLC motors to better support dynamic deliveries. To apply MLC tracking, we updated the leaf and jaw positions every 40 ms according to the predicted tumour position [24], thus correcting for translational target motion.

Imaging
The phantom motion was continuously monitored using sagittal 2D cine-MR [24]. Images were acquired using a T1-weighted gradient-echo sequence at 4 Hz, with a 2.5Â2.5 mm 2 voxel size and 350Â350 mm 2 field of view. Target positions were estimated from the images using cross-correlation based template matching [29].

Latency & Prediction
End-to-end latency, defined as the time lag between a physical motion event and the MLC response [30,24,22], causes dosimetric errors during MRI-guided MLC tracking. To estimate the average end-to-end latency, the Quasar MRI 4D phantom with circular 10 mm target was programmed with sinusoidal CC motion (A = 10 mm, T = 4 s). A squared 5Â5 cm 2 aperture tracked the target based on MR derived positions. An integrated electronic portal imaging device panel detected the target and MLC-aperture positions. Based on the phase difference between the two sets of positions, the end-to-end latency was estimated [30,19]. We derived the end-to-end leaf (s measured leaf ) and jaw (s measured jaw ) latency independently. Because the phantom has only one translational motion axis and it does not fit in the bore transversely, s measured jaw was estimated by software-wise mapping the phantom's CC motion to left-right (LR) motion such that the tracking software assumed pure LR motion and moved the jaws in LR direction to track the motion.
A linear ridge regression predictor mitigated the end-to-end latency by using past tumour positions to predict the next position [24]. The small difference between s measured leaf and s measured jaw was not accounted for in the prediction.

Dosimetry setup
The dosimetric benefit of VMAT + MLC tracking was assessed by comparing static deliveries to deliveries with respiratory motion in three different experimental setups ( Fig. 1): Setup A: Setup A (Fig. 1i) used the Quasar phantom with gafchromic EBT-3 or EBT-XD film. This setup was previously used for IMRT + MLC tracking [24]. We programmed the Quasar with a VMAT combined with MLC tracking on an MR-linac subject-recorded motion trace containing respiratory CC motion (A=11 mm, f =0.33 Hz, drift=0.3 mm/min). The phantom contained a film insert with 3 cm spherical target (GTV) that was positioned centrally or peripherally (10 cm off-center) in a water-filled body oval. Recorded 2D cine MRIs used for MLC tracking and the current positions simultaneously reported by the phantom (phantom reported positions) were saved for later use in Setup B. Setup B: Setup B (Fig. 1iii) used the Delta 4 phantom [31] with research software enabling VMAT support and time-resolved data export/input. It was longitudinally positioned in the bore either centrally or with a lateral offset, depending on the lateral position of the tumour. Two orthogonal planes (coronal and sagittal) filled with diodes measure dose every 25 ms. Diodes in the central 6Â6 cm 2 are spaced 5 mm apart, with 10 mm spacing elsewhere. The phantom cannot move and its electronics disallow real-time MR-imaging. Therefore, we used the Quasar in setup A to pre-record phantom motion using 2D-cine MRI at 4 Hz and simultaneously logged the phantom reported (groundtruth) positions on the same workstation. The positions were streamed to the tracking software as if they were acquired online.
Because the Delta 4 is immovable, the measured dose during MLC tracking is spread out relative to a static delivery. To retrospectively evaluate the notional VMAT + MLC tracking performance, we created a virtual HexaMotion platform. We virtually moved the phantom with the target by shifting the measured incremental dose (every 25 ms) by the same amount but in opposite direction as the target had moved w.r.t the isocenter. Dose shifts were carried out by re-gridding the data using cubic spline interpolation.

Setup C:
Setup C (Fig. 1v) was derived from Setup A, but with the phantom rotated by 20 around the anteroposterior axis to decompose the phantom motion into CC motion (A=10 mm) and LR motion (A=4 mm). This enables MLC tracking in both CC and LR direction.

Treatment planning
For our experiments, we used four patient lung SBRT plans with a central lung tumour and eight phantom plans for both central and peripheral indications. Planning CTs were acquired on the Brilliance Big Bore CT scanner (Philips Medical Systems, Best, The Netherlands). Patients underwent 4D-CT imaging from which an averaged 3D-CT image was reconstructed, while the Quasar phantom was imaged using 3D-CT. For each patient an ITV was delineated by a radiation oncologist and an isotropic 3 mm ITV-to-PTV margin was used following the institutional practice. In the phantom, the target was delineated as GTV and an isotropic 3 mm GTV-to-PTV margin was used. Additionally, an oesophagus, bronchus and aorta were delineated as OAR for central phantom plans. No OARs were delineated for the peripheral phantom plans. For each patient case, an 8Â7.5 Gy VMAT plan was created by adapting the patient's clinically approved lung SBRT IMRT plan to a VMAT plan (Fig. 2) in research Monaco 5.51.10 (Elekta AB, Stockholm, Sweden). For the phantom we created four VMAT plans for both experimental setup A and C by adapting the clinical lung SBRT template: two central 8Â7.5 Gy two-arc plans with and without OAR, and two peripheral 3Â18 Gy and 1Â34 Gy four-arc plans. Setup B used the same plans as setup A, but calculated for the Delta 4 . Following our clinical template, in all plans the maximum dose in the GTV was limited to 145% of the prescription dose (PD). Retrospective MU scaling was applied for all plans to provide a PTV coverage of 95% of the PD. Gantry angles 5-25 were excluded from the arcs, to avoid irradiating the cryostat pipe. Plans were calculated with 3 mm grid size and 8% Monte Carlo uncertainty per segment, following our current clinical practice.

Dosimetric evaluation
The irradiated EBT-3 and EBT-XD films were scanned and digitized with an Epson Expression 10000XL flatbed scanner (Seiko Epson Corp, Nagano, Japan). To analyse the films we used inhouse developed software. The films were semi-automatically registered with a point matching algorithm, based on three indents created by the phantom's film cassette. A 2%/1 mm local gammaanalysis evaluated the correspondence between static and MLC tracking deliveries for film and Delta 4 measurements. Only dose values with >10% PD were analysed, to reduce calibration induced uncertainties. For the film dosimetry, the GTV-coverage was quantified using dose area histograms (DAHs).

Plan deliverability calculation
We determined if the leaf and jaw motion of the VMAT delivery might be affected by the superimposed MLC tracking motion. The leaf/jaw speeds necessary for VMAT were based on the static delivery log files. Only segment shaping leaves were included, to avoid underestimation of the speeds due to static leaves outside the seg-ment shape. Additional leaf/jaw speeds necessary to track CC motion in setup A&B and CC + LR motion in setup C were based on the motion trajectory. The required leaf/jaw speeds were expressed as the near-maximum (95 th percentile) and the maximum (99 th percentile) speed.  Fig. 3 shows an overview of the dosimetric results of setup A&B for the 1Â34 Gy delivery. The dose difference maps emphasize that without MLC tracking the delivered dose differs from a static delivery, while the MLC tracking case is very similar. The delineated target in the dose maps (Fig. 3A) shows that without MLC tracking, parts of the GTV drift outside the prescription iso-dose line, indicating underdosage of the cranial GTV edge. This is supported by VMAT combined with MLC tracking on an MR-linac the DAH and the dose profiles. The time-resolved dose graphs (Fig. 3B) show the dose of a single diode. Initially, it receives the same dose in all three cases. After about 500 s the no MLC tracking case starts receiving less dose, which coincides with increased drift in the motion trace. Fig. 4 shows similar results for patient 1, where the dose with MLC tracking agrees well with the static dose, while without MLC tracking the dose differs. Fig. 5 shows the results for a 1Â34 Gy delivery of setup C. Here, MLC tracking and static dose maps also agree well, while without MLC tracking this agreement markedly deteriorates. The delineated target in the dose maps, the DAH and the line profiles show that MLC tracking effectively avoids underdosage of the GTV. Table 1 summarizes the gamma pass-rates and DAH metrics in the different setups. MLC tracking yields gamma pass-rates of 88-100% for film measurements in setup A&C and 92-100% for the Delta 4 in setup B. The peripheral plans have lower pass-rates without MLC tracking (22-38%) than the central deliveries (34-77%). These deliveries are longer and accumulate more drift. The gamma pass-rates for the phantom plans are similar to the pass-rates of the patient plans. The DAH results in Table 1 show similar GTV and PTV coverage for the static and MLC tracking cases, with dose differences within the GTV ranging between 0-15 cGy (8Â7.5 Gy), 5-102 cGy (3x18 Gy), 26-176 cGy (1Â34 Gy) for the D 98% , D 50% , and D 2% GTV dose. Without MLC tracking, the GTV D 98% reduces by 25-55 cGy (8Â7.5 Gy), 254-378 cGy (3Â18 Gy), and 575-645 cGy (1Â34 Gy), resulting for the peripheral cases in severe underdosage. Both in setup A&C, the PTV D 95% benefits from MLC tracking. Again the peripheral cases benefit most, as MLC tracking is needed there to achieve sufficient coverage. The GTV D 50% and D 2% were similar to the static case.

Discussion
In this study, we demonstrate the first experimental setup that combines VMAT with MRI-guided MLC tracking on an MR-linac. For a range of lung SBRT fractionation schemes, we demonstrate with our novel phantom setups that MLC tracking excellently compensates for respiratory motion during VMAT resulting in similar dose distributions and target coverage as during a static delivery.
The end-to-end leaf latency (328.05AE3.78 ms) was in line with previously reported values on Unity [24,22]. We found s measured jaw to be very similar to s measured leaf . We initially expected more latency for the heavier jaws, but because their motor controller compensates for their weight, the latency is similar. The predictor effectively mitigated the end-to-end latency. Because s measured jaw is lower than s measured leaf , we slightly over predicted for s measured jaw , leaving a negligible residual end-to-end latency of 34.86(AE4.58) ms. To deliver our VMAT plans, we required maximum MLC speeds of 12.1 cm/s. MLC tracking required additional speeds of maximally 1.48 cm/s. With these maximum speeds as determined, the MLC was still able to follow the combined VMAT and MLC tracking motion. We introduced three experimental setups to show the dosimetric benefits of MLC tracking during VMAT for lung SBRT. The results prove that VMAT benefits from MLC tracking, providing dose distributions very similar to the corresponding static delivery with local gamma pass-rates >88%. These values are in line with previously reported pass-rates for IMRT + MLC tracking [24]. For our motion trajectory, MLC tracking mainly benefits the coverage of the target periphery. Lung tumours may experience a baseline drift of up to 0.9 mm/min [32], meaning that they can easily drift outside the  Gy)  Static  -1959  2206  2404  1812  --1871  2110  2334  1761  No MLC tracking  33  1581  2130  2317  1395  38  33  1617  2057  2238  1461  MLC tracking  88  1930  2144  2302  1785  99  92  1908  2146  2329  1784  (1Â34 Gy)  Static  -3543  3960  4239  3292  --3845  4252  4626  3614  No MLC tracking  28  2898  4006  4359  2539  22  30  3270  4206  4444  2942  MLC tracking  93  3610  4025  4362  3333  100  92  3794  4190  4450  3588 3 mm GTV-to-PTV margin during longer treatments. In our experiments, the central phantom deliveries were relatively short deliveries of only 7 min resulting in little accumulated drift. This means that during the central deliveries the target did not drift out of the PTV margin, which explains that also without MLC tracking both the GTV and PTV coverage were hardly affected. This is also reflected in the relatively high Delta 4 gamma-pass rate, where the motion was too subtle considering the relatively coarse diode spacing. For the higher dose peripheral deliveries, combining VMAT with MLC tracking is crucial to restore target coverage, as these deliveries are long due to the relatively low dose rate of Unity (425 MU/min). A limitation of our dosimetric evaluation is that experimental setups A&C only evaluate dose in a single coronal plane intersecting the center of the target, while neglecting the rest of the target. The Delta 4 in setup B expands this evaluation from 2D to pseudo-3D by adding a sagittal plane to the evaluation. The results demonstrated that the effect of MLC tracking is identical in both planes, confirming the benefits of MLC tracking for VMAT. The timeresolved and motion included aspects of setup B, can also be expanded to applications outside this study.
Other than restoring target coverage, MLC tracking also reduced hot spots during VMAT outside the GTV as visible in the dose difference maps. Especially in the 1Â34 Gy deliveries without MLC tracking, there is up to 15 Gy extra dose outside the GTV compared to the static case. These undesired high dose areas could damage surrounding tissue [2,12]. This is also important for central tumours, for which intensified hypofractionation is currently often not recommended because of toxicity risks [33]. The high precision of VMAT + MLC tracking could increase the feasibility of single fraction deliveries for more patients, while also increasing patient comfort with the possibility of free-breathing during treatment and no extended beam-on times (100% duty cycle). This is especially true for patients with substantial tumour motion.
Neither MLC tracking nor VMAT are in the current clinical MRlinac release. Although we demonstrated the technical feasibility of VMAT + MLC tracking on the MR-linac, further work is needed to develop an integrated clinical workflow. The clinical treatment planning (TPS) system needs to fully support VMAT on Unity, especially for daily plan adaptation. Our research TPS occasionally creates segments with very low dose rates, limiting the deliverability of a plan. It also creates closed segments where leaves have to fully close and open in two consecutive segments, which is difficult to track because of the required additional leaf speed. Additionally, the imaging and motion estimation workflow needs to be tailored to patients [34]. The current sequence only images a 2D sagittal plane, as it was developed for the Quasar. Monitoring throughplane motion is crucial in real patients to avoid losing the target when it moves out of plane. To avoid increasing the system latency with relatively slow 3D sequences, interleaved orthogonal 2D-cine MRIs can be used to track 3D tumour motion [35,36] combined with a prediction filter to mitigate latency [36].

Conclusions
We provided a first experimental demonstration of the technical feasibility of VMAT + MLC tracking on an MR-linac for several lung SBRT indications. MLC tracking maximizes the sparing of healthy tissue in the presence of respiratory motion during VMAT, while ensuring target coverage.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.