Frame‐based radiosurgery of multiple metastases using single‐isocenter volumetric modulated arc therapy technique

Abstract Single‐isocenter volumetric modulated arc therapy (VMAT) technique can provide stereotactic radiosurgery (SRS) treatment with improved delivery efficiency for treating multiple metastases. Nevertheless, planning is time consuming and verification of frame‐based SRS setup, especially at noncoplanar angles, can be challenging. We report on a single‐isocenter VMAT technique with a special focus on improving treatment workflow and delivery verification to exploit the minimized patient motion of the frame‐based SRS. We developed protocols for preplanning and verification for VMAT and evaluated them for ten patient cases. Preplans based on MRI were used to generate comparable treatment plans using CT taken on the day of treatment after frame placement. Target positioning accuracy was evaluated by stereoscopic in‐room kV imaging. Dosimetric accuracy of the noncoplanar plan delivery was validated using measurement‐guided 3D dose reconstruction as well as film‐based end‐to‐end test with a Rando phantom. Average absolute differences of homogeneity indices, conformity indices, and V12Gy between MR preplans and CT‐based plans were within 5%. In‐room imaging positioning accuracy of 0.4 mm was verified to be independent of the distance to the isocenter. For treatment verification, average local and global passing rates of the 3D gamma (1 mm, 3%) were 86% and 99%, respectively. D99 values were matched within 5% for individual target structures (>0.5 cc). Additional film analysis confirmed dosimetric accuracy for small targets that had large verification errors in the 3D dose reconstruction. Our results suggest that the advantages of frame‐based SRS and noncoplanar single‐isocenter VMAT technique can be combined for efficient and accurate treatment of patients with multiple metastases.

accuracy. [1][2][3] These developments resulted in a paradigm shift toward definitively treating an ever increasing number of targets with radiosurgery. Treatment of multiple metastases, with traditional linac-based SRS approach still presents challenges due to significant effort in planning and delivering treatments to multiple isocenters.
Recent studies demonstrated that a single-isocenter technique combined with volumetric modulated arc therapy (VMAT) optimization for planning could be used to generate clinically acceptable plans that were comparable to the conventional multiisocenter approach. [4][5][6][7][8] The single-isocenter VMAT technique substantially lowers the beam-on time without compromising plan qualities such as dose fall-off and conformity. 6,9,10 In particular, the VMAT solution by Varian known as RapidArc was used by several investigators to achieve highly conformal dose distributions with optimal treatment delivery efficiency for multiple lesions.
Since VMAT planning is rather complex with multiple targets and organs at risk for optimization, it typically uses a workflow that employs a mask immobilization and CT simulation followed by several days of planning before the patient is treated. The frameless technique typically involves 1-3 mm planning target volume (PTV) margin. [11][12][13] In particular, rotational errors for targets at distances away from the isocenter would negatively affect the treatment delivery accuracy unless an additional margin is added due to the less effective immobilization and setup with a nonrigid mask system. [14][15][16] Also, accurately positioning the patient for treatment setup for all treatment couch angles becomes the challenge for this approach since multiple targets are treated with one isocenter. A more rigid patient immobilization and setup afforded by traditional invasive stereotactic frame would minimize any potential patient motion and eliminate rotational misalignments during treatment increasing the effectiveness of single isocenter VMAT SRS for multiple targets. This requires the complex VMAT planning and QA process to be carried out in a same day procedure of Computed Tomography (CT) simulation and treatment with less than 8 h of time window. Therefore, a preplanning is crucial to frame-based VMAT SRS to address planning challenges before the treatment day.
We devised a practical approach to improve workflow and also established the delivery accuracy of single isocenter VMAT for frame-based SRS of multiple metastases. We utilized Magnetic Resonance Imaging (MRI) scans of SRS patients, which are acquired 2-3 days before the frame placement for target definition, preplanning, and treatment optimization. We also validated the spatial accuracy of in-room kV imaging system combined with 6 degrees of freedom couch to be insensitive to the distance between the isocenter and each of multiple targets, and minimized the delivery time of noncoplanar beams using the image-based target positioning. The performance of patient-specific treatment verification was evaluated using patient geometry dose reconstructed from diode detector array measurements. This allowed dosimetric analysis for individual targets as well as 3D gamma evaluation for the acceptability of the entire plan.

| METHODS
RapidArc plans were prepared using Varian External Beam Planning version 13.7 and delivered on a TrueBeam radiotherapy system equipped with a Millenium 120 MLC (Varian, Palo Alto, CA, USA).
We evaluated patient-specific preplanning for ten patient cases with 3-9 intracranial lesions, 57 targets in total, which were treated at our institution over the past 2 yr as described below. Preplans were generated using the body contours and target volumes as defined in axial MR images. For the calculation of dose, Hounsfield unit of zero was assigned to the MR body contour. Target volumes ranged 0.03-20 cc with the prescription doses of 14-18 Gy. All plans had 360degree axial arcs and 180-degree vertex arcs at two couch angles.
Each arc was paired with one having an opposite gantry rotation and an orthogonal collimator angle. The plans were generated with the 6-MV flattening filter free beam at a dose rate of 1400 monitor units (MU) per minute using Photon Optimizer, Smart LMC, and Anisotropic Analytical Algorithm with 1.0-mm grid. Normal tissue objective was turned on and concentric optimization structures were created surrounding each target to maximize dose fall-off and minimize normal brain dose at V12Gy level. 6 Optimization objectives were adjusted to cover 99% of the gross target volume with the prescription dose while limiting the dose to any critical organs near the target volumes. Tumor volumes were primarily defined on the MR imaging which was always acquired within less than a week of the SRS delivery. On the SRS procedure day, MR information was registered with the spatially accurate CT which is localized in the stereotactic reference system to ensure accurate target localization. Then, the plan was re-optimized using the beam settings and constraints set retrieved from the preplan. With the same target volume coverage as that of the preplans, homogeneity indices (HI, maximum dose/prescription), conformity indices (CI, ICRU 62 definition), 18 and V12Gy were compared to evaluate the reproducibility of the plan quality. The MR images had slice thickness and in-plane resolution values ranging from 0.4 to 1.4 mm, while the SRS planning CT scans had isotropic resolution of 0.7 mm.
The treatment delivery used the PerfectPitch 6 degrees of freedom couch on TrueBeam combined with in-room kV x-ray imaging of ExacTrac system (Brainlab, Munich, Germany) for target positioning. ExacTrac x-ray positioning tolerance was set to the minimum (0.2 mm), and the "reference star" array was used to relay the couch correction information. The robotic positioning performance of the system was evaluated using orthogonal MV images of a polystyrene phantom with five imbedded radiopaque markers at arbitrary distances from the isocenter. The phantom had metallic wires wound in an arbitrary shape to drive the image guidance. The Digital Megavolt Imager of TrueBeam acquires 1280 × 1280 pixels over 43 cm × 43 cm field size, which defines 0.22-mm resolution with the source-todetector distance of 150 cm. The positioning error was defined as the distance discrepancies of the centers of BB's between the digitally reconstructed radiography (DRR) and the direct imaging with the therapeutic MV beams. 19 The distances were measured using the Varian Offline Review tools.
Dosimetric accuracy of the RapidArc plans was verified using a measurement-guided 3D patient dose with 1-mm grid reconstructed by ArcCHECK 3D diode detector array (Sun Nuclear, Melbourne, FL, USA) and 3DVH version 3.2. Briefly, the entrance and exit absolute dose measurements are summed into a 3D composite dose within the cylindrical volume of the ArcCHECK, which is subsequently morphed onto the patient geometry. 20 The measurement required noncoplanar couch angles to be reset to 0°, but the software reconstructed the 3D dose distribution of the original noncoplanar configuration from the coplanar measurement. Error specificity of the verification was evaluated using selected treatment plans cast on a Rando head phantom. Dosimetric errors between calculated plans and delivery measurements were induced by applying −6% to +6% normalizations to the intended plans. For example, dose measurement of an 18-Gy plan underwent verification analysis using the original plan (0% error) as well as plans normalized from the original treatment plan to deliver 16.9, 17.5, 18.5, and 19.1 Gy to the target volume, which correspond to −6%, −3%, 3%, and 6% induced errors, respectively. Also, to further evaluate the target size dependency of the error specificity, we prepared treatment plans treating four structure sets of Rando phantom: CT0.1cc, CT0.5cc, CT1cc, and CT5cc.
Each structure set had four spherical targets of the same size as denoted in the label.
We carried out the coplanar verification measurements for the ten patients, and evaluated 3D gamma passing rates at 1-mm distance and 3% dose difference with a 15% threshold for each patient plan as well as the D99 (dose that covers 99% of the PTV) for each target. We also randomly selected one patient plan and performed further film analysis to further study the dose statistics reported by 3DVH calculations. The corresponding plan was transferred to the Rando phantom with an adjustment of isocenter so that the target of evaluation falls on embedded Gafchromic EBT3 films. In order to do the film analysis, we normalized the prescription dose so that delivered dose could be appropriately assessed in the calibrated range of the film (<10 Gy). We followed the same imaging and positioning protocol for the film dosimetry as for patients, that is, delivering the noncoplanar "treatment" plan. Film dosimetry was performed using a home-built python script following the suggestions of previous studies. 21,22 Briefly, an EPSON Expression 10000XL scanner was used in 48-bit RGB mode at a resolution of 150 dpi, 24-h after irradiation. We used 8 cm × 8 cm films with a consistent orientation, and placed them at a reproducible central location of the scanner to minimize the lateral dependence artifacts.
We calibrated the optical density range 0-10 Gy using a weighted sum of the three channels of red, green and blue. Institutional Review Board approval was obtained for this study.

| RESULTS
The clinical plan qualities for all MR-based preplans and stereotactic CT plans were comparable in terms of target coverage statistics and normal tissue dose. Table 1 lists median and range of the dosimetric indices for the 57 targets in ten patients in this study. Homogeneity and conformity were evaluated for each target, and normal brain volume receiving 12 Gy (V12Gy) was calculated for each patient. Figure 1 shows the scatter plots of homogeneity index (HI), conformity index (CI), and noninvolved brain V12Gy. Overall, the target HI and CI as well as the normal brain V12Gy varied depending on the size, shape, and number of target volumes, but the CT plans reproduced similar values as those calculated with MR preplans with average absolute differences of 5.2%, 3.6%, and 2.1%, respectively. Figure 2 shows an example of dose distributions at an axial plane that intersects three target volumes (red). The shape of isodose lines of the MR preplan were closely matched to those of the CT plan. In particular, as one of the target volumes was near the optic nerve, the constraint set, and the prescription dose were customized to protect the critical organ. The two bottom panels show close-up views of the planar dose distributions for the optic nerve and the nearby target.     Figure 6 shows verification results of the ten patients. Average local and global passing rates of the 3D gamma [ Fig. 6(a)] were 86.2% and 99.0% with the standard deviations of 6.0% and 1.6%, respectively. Two patients had local gamma passing rates lower than 80%. There were 15 targets in them with ten targets less than 0.5 cc, and four small targets had D99 discrepancies greater than 5%. Figures 6(b) and 6(c) show errors in D99 for the ten patients plotted against target size and the distance to isocenter, respectively.
Overall, large errors reported by 3DVH verification were associated with small targets. For targets greater than 0.5 cc, D99 values agreed with the treatment plan within 5%. No overall trends were found with the distance to isocenter (r 2 < 0.1).
The tenth patient in the left panel of Fig. 6 was selected for further investigation. This patient had eight targets ranging 0.1-8.4 cc, and five targets were less than 0.5 cc. We performed film-based end-to-end test for two targets. Their volumes were 0.4 and 5.2 cc, and the values of D99 reported by 3DVH were 11.2% and 0.3% higher than that of treatment planning system (TPS), respectively.   Therefore, it is critical to verify the spatial positioning accuracy with a minimum possible tolerance for a frame-based technique, and also to assess any variation with the distance to the isocenter if multiple targets are handled by a single isocenter. Figure 4 demonstrates that the positioning performance of the stereoscopic in-room kV imaging was independent of the distance between the target and the isocenter. Furthermore, the average discrepancy of 0.4 mm between the DRR and the MV imaging indicates the highest attainable spatial accuracy with our setup, which is presumably limited by the intrinsic accuracy of DRR generated by the CT with the minimum slice thickness of 0.7 mm.
Frame-based immobilization is a mature technique and no additional treatment margin is needed for a target treated at the isocenter. 24,25 We used a dedicated target-positioning box provided by the manufacturer of our SRS positioning system for the initial patient setup and quality assurance of our single-isocenter multiple target treatments. We verified the patient setup further with image-guidance and confirmed that the spatial accuracy was consistently maintained at distances away from the isocenter for all targets with our frame-based setup. We point out, however, if a clinic attempts to use frame only setup without any image guidance, a careful assessment of rotational error should be carried out to determine corresponding treatment margins due to any possible residual rotations away from the isocenter. Any rotational error, however, intrinsically penalizes the smaller initial PTV margin by requiring a relatively larger additional margin for the off-isocenter targets. 17 Addition of extra margins would defeat the purpose of the invasive stereotactic frame placement, and may lead to a detrimental reduction in therapeutic ratio by increasing the risk of necrosis of normal brain tissue. 23 Volumetric modulated arc therapy involves delivery of inversely optimized dynamic MLC while gantry is in motion, which requires a patient-specific verification measurement. There are several conditions pertinent to this treatment modality. First, the verification needs to be performed within the limited time window of the frame-based workflow. For SRS, it is desirable to use measurement devices that would support a gamma analysis with at least 1-mm tolerance level. On the other hand, as discussed below, gamma passing rates may be suboptimal to assess a plan treating multiple metastases. Using ArcCHECK and 3DVH, we performed 3D gamma analysis in the patient geometry as well as dosimetric evaluation for individual target volumes.
The responses of 3DVH to the induced errors as shown in the phantom study (Fig. 5) demonstrates the complexity associated with the plan verification. Global gamma had poor specificity to the induced delivery errors, and was not acceptable for our purpose.
Local gamma analyses were compounded by target size dependency.
Using passing rates alone, it would be difficult to identify a potential delivery error with the realistic cases involving different target sizes. scatters. The magnitudes of these errors were independent of the distance to isocenter (Fig. 6c), which is consistent with the positioning accuracy analysis in Fig. 4. Instead, they were strongly correlated to the target sizes as shown in Fig. 6(b). The independent film measurements (Fig. 7) confirmed that these errors were not real, and they may represent the inherent limitation of the ArcCHECK/3DVH performance for small targets.

| CONCLUSION
We devised a practical approach that can combine advantages of frame-based SRS with noncoplanar single-isocenter VMAT in treating multiple metastases. MRI preplanning using VMAT SRS technique and plan QA performed with ArcCHECK/3DVH allowed for efficient planning and accurate delivery verification within the limited time window.
ExacTrac in-room kV imaging expedited noncoplanar patient setup with the invariant spatial accuracy across multiple targets.

ACKNOWLEDG MENT
Dr. Ahn wishes to thank Dr. Howard J. Halpern who encouraged writing up this work.

CONFLI CT OF INTEREST
No conflict of Interest.