Feasibility of virtual starshot analysis providing submillimeter radiation isocenter accuracy: A long‐term multi‐institutional analysis

Abstract Purpose We developed a technique to calculate the offset between room lasers and the radiation isocenter using a digital Winston–Lutz (WL) test with a starshot technique. We have performed isocenter localization quality assurance (QA) with submillimeter accuracy for a long period. Here we evaluated the feasibility and accuracy of this virtual starshot (VS) analysis for isocenter localization QA. Methods A 6‐MV photon beam with a square multileaf collimator field was used to irradiate a WL sphere positioned at the intersection of the room lasers. Images were acquired using an electronic portal imaging device. A four‐field WL test was performed, and the path of each beam was calculated from the offset between the beam and sphere. Virtual starshot analysis was used to analyze the radiation isocenter, which calculates the center of the beam paths by using a least‐squares method, similar to the starshot analysis. Then, eight coplanar and 12 noncoplanar beams were irradiated to evaluate isocenter localization accuracy. Results Several VS analyses, using different WL spheres, were performed at three institutions, and the calculated accuracies were within 0.1 mm at all institutions. Long‐term analysis showed that the isocenter localization accuracy was appropriately managed with three‐dimensional accuracy within ± 0.5 mm for 90 months after the first laser adjustments. The offset between each beam and the room laser was within 0.6 mm and within 1.0 mm for eight coplanar and 12 noncoplanar beams, respectively, for 90 months. Cone‐beam computed tomography images, acquired after verification beams, showed that the offset between the radiation isocenter and the imaging center was within 0.66 mm for 90 months. The isocenter localization accuracy within 1 mm was kept for long period at other four institutions. Conclusions Long‐term analysis showed the feasibility of VS analysis for isocenter localization QA, including room laser re‐alignment, noncoplanar irradiation verification, and image guidance accuracy.

Such advanced techniques require high accuracy in mechanical structures, image-guided radiotherapy (IGRT), patient immobilization, and beam modeling of treatment planning systems (TPS) for small-field dosimetry. 5,6 The American Association of Physicists in Medicine, Task Group 142, recommends an accuracy of < ±1 mm of the localizing lasers as well as the imaging and treatment coordinate coincidence. 7 One of the major techniques for evaluating isocenter localization is the Winston-Lutz (WL) test, which uses a small sphere and a piece of film and evaluates the targeting accuracy of irradiation by checking whether the sphere is projected inside the radiation field. 8 Although the original WL test used four oblique gantry angles, with and without couch rotation, various patterns have been reported 9,10 . This technique is useful for quality assurance (QA) of the SRS for brain tumors, which often use noncoplanar beams to achieve highly conformal dose distribution.
Recently, WL tests using an electronic portal imaging device (EPID) have been reported. [10][11][12] The EPID generates digital image data and enables quick and quantitative evaluation of the WL tests.
Several studies have reported an algorithm that calculates the isocenter location with submillimeter accuracy; others have stated that their digital WL tests can be used to re-align the room lasers to target the radiation isocenter. [11][12][13][14][15] Even if the offset between the room lasers and radiation isocenter is calculated with submillimeter accuracy, however, it is very difficult to adjust the room laser localization to the radiation isocenter with submillimeter accuracy because the translations, rotations, and tilts of the lasers equipped in a radiotherapy treatment room are usually controlled manually. We developed an analysis technique to calculate the offset between the room lasers and radiation isocenter using a WL test and a starshot technique. Although previous studies 9, 12 have shown techniques to localize the isocenter accuracy using the WL test, they did not report on the routine use of their techniques for long term. Our institution used this "virtual starshot" analysis (VS analysis) as routine isocenter localization QA for a long period. In addition, we collected the VS analysis data from multiple institutions where this method was used periodically. Here we analyzed the long-term data of multiple institutions to evaluate the feasibility and accuracy of VS analysis.

2.A | Image acquisitions
In this study, images for the WL test data were collected from six institutions (A-F). VS analysis was routinely conducted at institution A and C-F. The linacs and spheres used for the WL test are listed in Table 1. The external visible marks on the sphere phantoms were aligned with the room lasers. Here, two centers were defined: the sphere center and radiation isocenter. Images were acquired using an EPID. Most institutions used the field size of 20 × 20 mm 2 , whereas only institution A used the size of 10 × 10 mm 2 because this institution conducted brain SRS using 2.5-mm MLC leaf width.
Because the MLC's nominal positional accuracy was better than that of the jaws, the field size was defined by MLC. The collimator angle was set to 90°for all beams to minimize the effects of gravity, although patients were treated with various collimator angles optimized for each plan. Figure 1(b) shows a WL test rod containing a 5mm tungsten sphere (Taisei Medical, Osaka, Japan), which was manufactured as an attachment for Iso-Align (CIVCO Medical Solutions, Orange City, IA). First, a stainless rod with cross-hair lines (shown in the right-bottom window) was used to adjust the position to the room laser, and then the rod was exchanged with the WL test rod.
Figures 1(c) and 1(e) shows hand-made acrylic rods with a 3-mm steel sphere and a 5-mm tungsten sphere, respectively. The spheres were painted white to improve the visibility of the laser projected on it.

2.B | Winston-Lutz test
The acquired images were analyzed using Akilles RT software (RADLab, Osaka, Japan). A flowchart and calculation algorithm of the WL analysis are shown in Fig. 2. First, the pixel size recorded in the DICOM file header was corrected by source-imager-distance, and the images were resized to twice as large using linear interpolation to detect the sphere and field edges smoothly; the pixel size was 0.131 mm at the isocenter plane after correction for EPID with original pixel size of 0.392 mm. This process may lead to uncertainty of the analysis, although the impact is smaller than pixel size. To determine a threshold for detecting the sphere, an initial point was manually set. From the initial point, the pixel values were scanned radially every 2°and line profiles were acquired. To obtain the contour of AKINO ET AL. Here, PV max and PV initial represent the global maximum pixel value and initial pixel value, respectively. The sphere centroid was defined at the intersection of the room lasers. To reduce the uncertainty owing to the manual procedure of setting initial point, the calculated sphere center was set as the initial point, and the analysis was repeated. The contour of the radiation field was also extracted from the radial profiles using a threshold value of [0.3 × PV max + 0.7 × PV min ] and the radiation field centroid was calculated. Here, PV min represents global minimum pixel value. Then, the offsets between the radiation field and sphere represent the relative displacements of the radiation beam from the room laser (δ Beam, Laser ).
Another contains eight coplanar beams with 45°of gantry angle intervals and 12 noncoplanar beams. For noncoplanar beams, the couch rotation angles were 45°, 90°, 270°, or 315°. For each couch rotation angle, three gantry angles were tested: 0°, 30°, and 210°or 0°, 150°, and 330°depending on the clearance between the couch basement and the linac head. Figure 3(a) shows a flowchart of the VS analysis procedure. First, the WL test images were acquired, and the δ Beam, Laser value was calculated for each image, as described in  vertical (Y) coordinates of these two points were calculated as follows:

2.D | Validation
To validate the technique presented in this study, we performed VS analysis at institution A-C with various devices, as illustrated in  Table 2 shows the results of the four-field VS analysis performed at

| DISCUSSION
In this study, we reported a technique to determine the radiation isocenter position using VS analysis. We also assessed the feasibility of VS analysis for adjusting the room lasers projecting to the radiation isocenter by using long-term data. The technique presented in this study is simple, quantitative, free of inter-observer variation, and free of uncertainties because of manual processes such as marking for film-based starshot tests. Image acquisitions were performed on the treatment plans of a virtual patient for QA. Therefore, irradiations and DICOM export can be easily performed. Because we checked the submillimeter accuracy of the couch movement of the TrueBeam at institutions A-C, the couch offsets after four-field VS analysis can be digitally and remotely conducted. Therefore, the procedure, including the WL phantom setup, irradiation of the four-field beams, DICOM export, VS analysis, couch offset, re-irradiation of the four-field beams, and second VS analysis can be completed within 20 min, although the verification beam irradiation takes a little longer because of the couch rotations for noncoplanar beams.  or all beams to avoid the gravity effects with gantry rotations.
Because the jaw field size was slightly larger than the MLC field size, the center and width of the coplanar beams on the transverse plane were defined by the four-times of the MLC leaf width, which results in very small uncertainties. To compensate for the uncertainties due to the tongue-and-groove effects, both 90°and 270°should be used although the measurement time will double. For VS analysis using coplanar beams, the gantry angles must be equally distributed because the center position was calculated using the least-squares method. In this study, both four-field and eight-field VS analyses used equally distributed gantry angles. Therefore, the results were expected to consider gantry sagging and misalignment of the gantry rotation axis. The gantry and couch angle of the noncoplanar beams did not need to be equally distributed because the values were not used to calculate the radiation isocenter position.
The technique presented in this study includes the following limitations. First, this method evaluates only the combination of gantry and couch angles with which the images can be acquired using EPID.
For SRS and SRT for brain metastases, the dynamic conformal arc or VMAT beams are often used with couch rotations. In many such cases, however, the EPID cannot be used because the panel collides with the couch. In this study, we tested the gantry angles of 0°, 30°, and 150°for a 270°couch angle and 0°, 210°, and 330°for a couch angle of 90°. Therefore, we partly checked the noncoplanar situations. Second, the WL sphere was remotely moved using the True-Beam couch movements after the initial four-field VS analysis.
According to the vendor-provided specification, the spatial translational accuracy of the treatment couch is ≤ 0.5 mm, but it appeared to be ≤ 0.2 mm in our experience of a monthly mechanical QA.
However, such uncertainty will still be smaller than that of the manual adjustments of the sphere position. If the sphere is located at an inappropriate position, it can be detected by the second four-field VS analysis. In our method, the room laser position was defined as the centroid of the sphere. Because the WL sphere is positioned manually by the staff, the initial room laser position of the analysis may be uncertain. If repositioning is conducted after VS analysis, the final position of the sphere will not remain uncertain, although manual adjustment of the room laser will result in additional uncertainty.
These will depend on the WL test devices and on the flexibility of the laser adjustments. Third, the analysis of our WL tests was not fully automated. This algorithm requires manually setting the initial position around the sphere center because the software was designed not only for EPID but also for starshot and WL tests using radiochromic film. This manual process was needed because the optimal threshold for detecting the sphere may vary depending on the sphere size and material. For example, institution C used a handmade WL rod with a 3-mm steel sphere whose radiopacity was lower than that of the tungsten sphere used at other institutions.
However, our algorithm successfully analyzed the WL tests for all three spheres. Because the effects of the manual process on the results were negligible (<0.1 mm), all images usually can be simultaneously analyzed within a few seconds by using the same initial position selected for one of the images. At institution C, however, we created another hand-made WL rod with a 5-mm tungsten sphere and obtained clearer images (data not shown). Creating a hand-made WL rod and a sphere with highly radiopaque material will provide images with better contrast.

CONFLI CT OF INTEREST
The authors YA and HS developed Akilles RT, which is a software for QA and research purposes and is currently a commercial software.

R E F E R E N C E S
1. Andrews DW, Scott CB, Sperduto PW, et al. Whole brain radiation therapy with or without stereotactic radiosurgery boost for patients with one to three brain metastases: phase III results of the RTOG 9508 randomised trial. Lancet. 2004;363:1665-1672.