Stereotactic radiosurgery with MLC‐defined arcs: Verification of dosimetry, spatial accuracy, and end‐to‐end tests

Abstract Purpose To measure dosimetric and spatial accuracy of stereotactic radiosurgery (SRS) delivered to targets as small as the trigeminal nerve (TN) using a standard external beam treatment planning system (TPS) and multileaf collimator‐(MLC) equipped linear accelerator without cones or other special attachments or modifications. Methods Dosimetric performance was assessed by comparing computed dose distributions to film measurements. Comparisons included the γ‐index, beam profiles, isodose lines, maximum dose, and spatial accuracy. Initially, single static 360° arcs of MLC‐shaped fields ranging from 1.6 × 5 to 30 × 30 mm2 were planned and delivered to an in‐house built block phantom having approximate dimensions of a human head. The phantom was equipped with markings that allowed accurate setup using planar kV images. Couch walkout during multiple‐arc treatments was investigated by tracking a ball pointer, initially positioned at cone beam computed tomography (CBCT) isocenter, as the couch was rotated. Tracks were mapped with no load and a 90 kg stack of plastic plates simulating patient treatment. The dosimetric effect of walkout was assessed computationally by comparing test plans that corrected for walkout to plans that neglected walkout. The plans involved nine 160° arcs of 2.4 × 5 mm2 fields applied at six different couch angles. For end‐to‐end tests that included CT simulation, target contouring, planning, and delivery, a cylindrical phantom mimicking a 3 mm lesion was constructed and irradiated with the nine‐arc regimen. The phantom, lacking markings as setup aids was positioned under CBCT guidance by registering its surface and internal structures with CTs from simulation. Radiochromic film passing through the target center was inserted parallel to the coronal and the sagittal plane for assessment of spatial and dosimetric accuracy. Results In the single‐arc block phantom tests computed maximum doses of all field sizes agreed with measurements within 2.4 ± 2.0%. Profile widths at 50% maximum agreed within 0.2 mm. The largest targeting error was 0.33 mm. The γ‐index (3%, 1 mm) averaged over 10 experiments was >1 in only 1% of pixels for field sizes up to 10 × 10 mm2 and rose to 4.4% as field size increased to 20 × 20 mm2. Table walkout was not affected by load. Walkout shifted the target up to 0.6 mm from CBCT isocenter but, according to computations shifted the dose cloud of the nine‐arc plan by only 0.16 mm. Film measurements verified the small dosimetric effect of walkout, allowing walkout to be neglected during planning and treatment. In the end‐to‐end tests average and maximum targeting errors were 0.30 ± 0.10 and 0.43 mm, respectively. Gamma analysis of coronal and sagittal dose distributions based on a 3%/0.3 mm agreement remained <1 at all pixels. To date, more than 50 functional SRS treatments using MLC‐shaped static field arcs have been delivered. Conclusion Stereotactic radiosurgery (SRS) can be planned and delivered on a standard linac without cones or other modifications with better than 0.5 mm spatial and 5% dosimetric accuracy.

nine-arc plan by only 0.16 mm. Film measurements verified the small dosimetric effect of walkout, allowing walkout to be neglected during planning and treatment.
In the end-to-end tests average and maximum targeting errors were 0.30 ± 0.10 and 0.43 mm, respectively. Gamma analysis of coronal and sagittal dose distributions based on a 3%/0.3 mm agreement remained <1 at all pixels. To date, more than 50 functional SRS treatments using MLC-shaped static field arcs have been delivered.
Conclusion: Stereotactic radiosurgery (SRS) can be planned and delivered on a standard linac without cones or other modifications with better than 0.5 mm spatial and 5% dosimetric accuracy.

K E Y W O R D S
end-to-end verification, linear accelerator, stereotactic radiosurgery, TGN

| INTRODUCTION
Stereotactic radiosurgery (SRS) is a valuable tool for the treatment of brain metastases, arterio-venous malformations (AVM), and functional brain conditions like trigeminal neuralgia (TN) and vestibular schwannoma. [1][2][3][4] It is typically administered with dedicated equipment like the GammaKnife ® (GK) or CyberKnife ® that is available only at a few facilities.
Considering how many linear accelerators (linac) are available for conventional radiotherapy, extending their use to SRS could greatly expand access to that modality. Linac-based SRS was introduced in the 1980s using standard treatment planning systems (TPS), occasionally supplemented by measured small-field dose distributions. X-ray jaw-defined fields were used for delivery. [5][6][7] Geometric and dosimetric accuracy was later augmented by in-house developed software and accessories, as well as by fine-tuning accelerators. 1,2,[8][9][10][11][12][13] As technology evolved more precise linacs were made and special accessories and TPSs for SRS became commercially available. 4,[14][15][16][17] However, at institutions with limited resources the extra effort of commissioning and maintaining any additional system can be prohibitive. Substantial research has been devoted in recent years to circumvent the need for special apparatus by testing the suitability of multileaf collimators (MLC) and standard TPSs for SRS.
Dosimetric accuracy of small MLC-shaped fields was investigated by numerous researchers. Poffenbarger et al 18 studied the performance of the Eclipse and iPlan TPSs for square fields having side lengths as small as 2.5 mm, shaped by the HD120 leaf MLC. Film data agreed within 5% with the TPS for field sizes 7.5 mm or greater, but for smaller fields Eclipse underestimated the dose by 11% or more.
Arcing fields showed similar behavior. Hrbacek et al 19 compared dosimetric parameters from Eclipse with measurements for stationary beams of 1 × 1 cm 2 and larger, delivered with a 120-leaf MLC to a water phantom. Agreement between measurement and experiment was considered acceptable. Audet et al 20  Collectively, the reports suggest that accurate small target SRS may be achievable with linacs and MLC-defined fields. However, the papers address individual aspects of SRS and on a variety of accelerators and TPSs, leaving concern if all components would work coherently together on a single system. This paper presents a comprehensive series of tests of a widely used combination of linac and TPS before it was put into clinical service for treating very small targets.
We consider only arcs since cranial SRS is commonly administered with rotating fields. Potential error sources are minimized by using the more recent Eclipse Version 13.6.30 and the smallest available grid spacing of 1.0 mm, an MLC with 2.5 mm wide leaves, fields that are at least two leaves wide and prevent leaf motion by applying static arcs.
The investigation consists of three basic parts. Firstly we tested if the TPS could accurately predict dose distributions of small MLC- shaped fields suitable for functional SRS (fSRS). Single arcs ranging from 1.6 × 5 to 30 × 30 mm 2 were planned and delivered to a novel high-precision block phantom containing radiochromic film in the coronal plane. The phantom had the approximate size of a human head and was equipped with markings that allowed better than 0.1 mm accurate setup on the accelerator by planar kV images. As a merit over commercial anthropomorphic phantoms that require setup based on CTs from simulation, our phantom isolated planning and delivery from potential upstream simulation errors.
In support of multiple non-coplanar arc treatments the second part of the project investigated couch walkout under rotation, with and without a 90 kg load. A novel spherical pointer was constructed for mapping the trajectory of a point, originally positioned at CBCT isocenter, as the couch is rotated. The dosimetric effect of the measured walkout was then explored using the TPS and measurements.
The third part of the investigation involved end-to-end tests in compliance with ASTRO recommendations. 26 These started with CT simulation of a cylindrical phantom having a 3 × 3 × 3 mm 3 target replicating a TN treatment. The target was contoured and a treatment involving multiple arcs at different couch angles planned and delivered. The phantom contained film in the coronal and sagittal planes and had no marks or other setup aids. Similar to a patient, it was positioned for treatment using the six-dimensional (6D) couch and automated matching of structural features. The level of agreement between planned and measured dose distribution was a gauge for expected clinical performance.
Consistency of the delivery method was examined by irradiating the block and the cylinder phantom 10 or more times during a 6month period. Performance of the block phantom and the cylindrical phantoms, in turn, was vetted by separate tests.
At the time of this writing static MLC-defined 2.1 × 5 mm 2 arcs at 5 equally spaced couch angles have been used to administer more than 50 rhizotomies of the TN and three thalamotomies under institutional review board (IRB) approved protocols. 27,28 In the TN treatments the target is the root entry zone, a maximum dose of 80 Gy is prescribed, and isocenter placed such that the 40 Gy surface abuts the brainstem. For thalamotomy the isocenter is placed at the ven- where. 29 The potential to generate specific non-spherical dose distributions is demonstrated in this paper by a nine-arc plan that closely resembles a GK with a 4 mm collimator.
In addition to its utility for fSRS with stationary arcs, our quality assurance (QA) method and block phantom have been used for more than 500 radiosurgeries delivered with VMAT and intensity modulated arc therapy (IMRT). These included primary brain tumors, metastases, and AVMs.

| MATERIALS AND METHODS
Throughout this paper we use the IEC 61217 coordinate convention.
For an observer standing at the foot of the treatment table and facing the gantry the positive x-, y-, and z-directions extend, respectively, left to right in the crossplane, toward the gantry along the caudal-cephalad direction in the in-plane, and toward the ceiling.
The couch angle is 270°when the couch is in the 9 o'clock position, and increases with counterclockwise (CCW) rotation to 360°/0°as the home position is approached, and further increases from 0°to 90°when the couch reaches the 3 o'clock position.
All measurements were done on an Edge accelerator equipped with a 120-leaf MLC (Varian Medical Systems, Palo Alto, CA). Dose distributions were computed on the Eclipse, Version 13.6.26 TPS with AAA algorithm provided by the same manufacturer.

2.A | Description of the block phantom
The phantom was made of polymethylmetacrylate (PMMA), measured 18 × 19 × 17 cm 3 , and consisted of two approximately equal parts that allowed insertion of GafChromic EBT-XD film (Ashland, Bridgewater, NJ) along the coronal plane [ Fig. 1(a)]. For film marking the phantom was equipped with precisely milled channels that served as needle guides.
To allow accurate positioning on the accelerator, the upper section of the phantom had a 5 cm radius circular groove engraved at the film plane that was visible on anterior-posterior (AP) kV setup images.
Matching the groove to a computer-generated circle centered at kV isocenter afforded phantom positioning to better than 0.1 mm along the coronal plane [ Fig. 1   fields. All irradiations were planned to deliver 8 Gy at isocenter, and were repeated 10 times at each field size.
The planning system was commissioned assuming a single point radiation source, 1.2% leaf transmission, and a dosimetric leaf gap of 0.86 mm, measured according to the sweeping gap method recommended by the manufacturer. Depth dose fractions and beam profiles were measured in a water tank covering the range of 3 × 3 to 40 × 40 cm 2 fields. The "golden beam data" currently provided by the manufacturer were not available at the time of commissioning.

2.D | Measurement of couch walkout
Even in a well aligned accelerator the couch axis may not pass exactly through isocenter and couch motion may not be perfectly smooth. 13,23 When the couch is rotated for multiple-arc treatments the target shifts from its original setup position, resulting in potentially inaccurate dose delivery.
To quantify walkout we built a novel pointer for mapping the trajectory of a target positioned at CBCT isocenter as the couch is rotated [ Fig. 2(a)]. The pointer consisted of a 25.4 mm diameter.
PMMA sphere with a concentric 6.35-mm tungsten carbide sphere and was mounted on micrometer-adjustable linear translation stages (Newport, Irvine, CA). The surface of the CBCT-imaged plastic ball was not affected by artifacts cast by the metal ball that was provided for Winston-Lutz (WL) tests. 9 We were able to position the pointer with better than 0.1 mm accuracy at CBCT isocenter by matching its image to computer-generated circles in the three princi-

2.E | Effect of couch walkout on dose distributions
The detrimental effect of walkout on the dose cloud of non-coplanar arcs was investigated theoretically and experimentally. In the former method, walkout at angles used in a test plan was inserted into the TPS as isocenter shifts. Dose profiles of the so generated plan were compared to profiles of a plan that neglected walkout. The test plan was based on the block phantom. It consisted of nine non-coplanar arcs of 2.4 × 5 mm 2 MLC-defined fields delivered at six different couch angles. Each arc covered 160° . Two arcs, 190°to 350°and   10°to 170°, were planned at table angles of 0°, 10°, and 350°while one arc was planned for 45°, 315°, and 90°couch rotation. Complete and 180°partial arcs were avoided to prevent hot spots at the anterior and posterior points of arc convergence.
Experimental testing involved planning the nine-arc treatment for the block phantom without application of isocenter shifts, and delivery without couch corrections for walkout. Since this procedure assumed a perfect couch in planning whereas delivery involved walkout, agreement between plan and experiment was expected to be worse than in the single-arc deliveries. However, based on the findings of the theoretical investigation, which suggested that couch walkout would have only a minor effect on the dose cloud, disagreement between plan and experiment should be only minor. The experiments were intended to verify that assumption.

2.F | Description of the cylinder phantom for end-to-end tests
The phantom for the end-to-end test is shown in Fig. 3 center of each section. When the two sections were joined the cavity simulated a 3 mm long 3.175 mm diameter lesion. Provisions were made for marking the film at the exact cavity center.

2.G | Accuracy of measurements with the cylinder phantom
These tests were designed to assess the accuracy of three-dimensional measurements made with the cylinder phantom. The phantom was small and light so that it could be mounted on precise linear translation stages that allowed phantom shifts at 0.01 mm accuracy.
Initially the phantom was set up at isocenter by registering CBCT images to planning CTs. Accurately known shifts were then applied using the micrometer drives without other changes to the setup, thus avoiding experimental errors that could be introduced if shifts were done by repeated phantom setup. Radiation was delivered at each position twice, with film positioned along the coronal and the sagittal plane, respectively. Single 360°arcs of a 2.4 × 5 mm 2 field were used.
Films were scanned and evaluated with ImageJ 1.48v, a program provided by the National Institute of Health. 31 Phantom positions with respect to the TIC were derived from the distance between the pin prick marking cavity center and the centroid of the 50% isodose line that represented TIC. Distances between measured phantom positions were considered as measured phantom shifts, and were compared to the known applied shifts.

2.H | End-to-end measurements with the cylinder phantom
The tests mimicked an intracranial treatment, starting at simulation followed by contouring, treatment planning, and delivery. Subsequent to simulation the target cavity was contoured and the nine-arc regimen consisting of 2.4 × 5 mm 2 arcs was planned to deliver 8 Gy at the cavity center. The electron density in the plan was set equal to PMMA density. Before irradiation, the air cavity was plugged with PMMA inserts to make it invisible on CBCT, mimicking a lesion identifiable on simulation CT but obscure to CBCT. The phantom was set up on the linac by matching CBCT images (head protocol, 100 kV, full rotation, 1 mm slices) to planning CTs of internal structures, similar to patient setup based on bony anatomy near the target. Automated registration and 6D couch movements were utilized, followed by a second CBCT to confirm accurate setup and, if necessary, small manual linear adjustments for optimal match. Radiation was delivered with marked films placed in the coronal plane and again with film in the sagittal plane. A third irradiation of unmarked film avoided dose errors in the vicinity of the pin prick.
Following scanning, films were evaluated with ImageJ. The distance between the centroid of the 50% isodose line and that of the prick mark at cavity center was the measure of spatial accuracy. All steps, including simulation, contouring, planning, and delivery were repeated ten times during a 6-month period to evaluate stability. A second, independent method of film evaluation was used in later tests. It was based on a modification of the previously described MatLab script and provided overlays between plan and delivery in the coronal and sagittal planes.     Table 2 for a range of field sizes. Data in rows 1-9 represents averages of 10 measurements. The γ-index is shown in the last T A B L E 1 Block phantom performance. Doses in Gy, distances in mm. The shift errors Δx and Δy are the differences between the individual measured shifts distances from one position to the next minus the respective known distances.  We experimentally verified the validity of the analysis by recording the trajectory of a pointer, originally positioned at CBCT isocenter as the couch was rotated. From the trajectory we computed v optimal and the shape of the wobble pattern that would result if the pointer was placed at the optimal positon. We applied v optimal to the pointer using the micrometer stages and repeated the trajectory measurement. All points of the so-found track agreed within better than 0.1 mm with the computed wobble pattern. Vertical excursions of the pointer, indicative of vertical couch walkout, were less than ±0.05 mm and considered negligible.

3.C | Couch walkout
The effect of load on walkout is shown in Fig. 6. In the first mea-

3.D | Effect of couch walkout on dose distributions
The couch walkout shown in Fig. 5(a) was applied in the nine-arc treatment plan for the block phantom as isocenter shifts of the respective arcs. We chose this trajectory because it was the largest one encountered during a 1-month monitoring period and therefore would show the largest dose errors. Figure 7 shows the effect of the couch deviations by comparing computed dose profiles that incorporate the shifts to profiles of a perfect couch (no wobble, couch axis agrees perfectly with CBCT isocenter).
According to the theoretical investigation couch walkout should have only a minor effect on the dose cloud. Based on that finding it would not be necessary to correct for walkout on our accelerator. We verified this hypothesis by planning a treatment with the nine-arc configuration without consideration of couch walkout and delivery with the couch that exhibited the measured walkout. A comparison between plan and measurement is shown in Figure 8 and  Table 4 compares measured phantom shifts to known shifts applied with micrometer-driven translation stages. Note the close agreement between applied and measured shift distances. It not only signifies an accurate phantom and experimental procedure but also a highly repeatable dose delivery by the accelerator. Any change in beam geometry between consecutive deliveries would appear as a discrepancy between applied and measured shifts.

3.E | Verification of cylinder phantom accuracy
3.F | End-to-end measurements with cylinder phantom Figure 9 shows In the clinic, the MLC-shaped fields were found practical in more than 50 fSRS treatments. We use 2.1 × 5 mm 2 arcs at five equally spaced couch angles to obtain near-spherical dose distributions. 29 A dose cloud resembling a GK with 4 mm collimator is presented in T A B L E 3 Quantitative comparison between planned and measured dose distributions of non-coplanar arcs delivered to the block phantom. Nine 160°arcs at six different couch angles were planned to deliver 8 Gy at isocenter using 10 MV flattening filter free beams. Data are averages of 10 measurements, defined as in the caption of Table 2.  23 and in our own work on the Edge.
We believe that the rotational application of radiation averages deviations at individual gantry angles, causing a small widening of beam profiles but having little effect on the center of the radiation cloud.
Our measured dose profiles matched the planned ones accurately, demonstrating that the effect is only minor.
The slightly larger targeting errors in the nine-arc delivery compared to single-arc delivery were likely caused by couch walkout.
Dose errors due to walkout could be eliminated by shifting the couch or the MLC as the couch is rotated. 1 Our initial concern about continued accuracy was largely dispelled by the QA tools provided by the manufacturer. Congruence of kV imaging isocenter and 6 MV TIC is assured by the automated IsoCal alignment procedure. 39 Congruence of the 10 MV FFF isocenter with imaging isocenter -we use 10 MV FFF for all SRS cases-is attained during accelerator installation by matching all photon beams to the 6 MV beam.
The recently introduced machine performance check (MPC) 42 Our QA method also provides for expedient tests following repairs or adjustment of critical components. For example, we remapped couch walkout after the linac was repositioned for closer match between couch axis and CBCT isocenter. According to our test, the adjustment moved the CBCT isocenter along the positive ydirection from its original position of 0.35 mm inferior of the couch axis to 0.15 mm superior. The engineers who had done the adjustment confirmed a 0.5 mm shift of the accelerator along the +y-direction, in agreement with our measurement. We also found that the realignment reduced maximum couch walkout by 0.15 mm.
Because of the heightened demands on accuracy in fSRS, we run IsoCal and MPC the day before treatment and perform WL tests at a multitude of gantry and couch angles. These are in addition to the daily WL spot checks done at a few selected gantry angles. We also measure the output of a single arc using a scintillator. While arc treatment with stationary MLCs would normally not require patientspecific QA checks, nevertheless we carry out such tests with the block phantom as another safeguard. Finally we want to point out that, while our TPS provides accurate output based on the original commissioning data, other TPSs may require adjustment of input data for optimal small-field calculations.