A novel method for monitoring the constancy of beam path accuracy in CyberKnife†

Abstract The aim of current work was to present a novel evaluation procedure implemented for checking the constancy of beam path accuracy of a CyberKnife system based on ArcCHECK. A tailor‐made Styrofoam with four implanted fiducial markers was adopted to enable the fiducial tracking during beam deliveries. A simple two‐field plan and an isocentric plan were created for determining the density override of ArcCHECK in MultiPlan and the constancy of beam path accuracy respectively. Correlation curves for all diodes involved in the study were obtained by analyzing the dose distributions calculated by MultiPlan after introducing position shifts in anteroposterior, superoinferior, and left–right directions. The ability of detecting systematic position error was also evaluated by changing the position of alignment center intentionally. The one standard deviation (SD) result for reproducibility test showed the RMS of 0.054 mm and the maximum of 0.263 mm, which was comparable to the machine self‐test result. The mean of absolute value of position errors in the constancy test was measured to 0.091 mm with a SD of 0.035 mm, while the root‐mean‐square was 0.127 mm with a SD of 0.034 mm. All introduced systematic position errors range from 0.3 to 2 mm were detected successfully. Efficient method for evaluating the constancy of beam path accuracy of CyberKnife has been developed and proven to be sensitive enough for detecting a systematic drift of robotic manipulator. Once the workflow is streamlined, our proposed method will be an effective and easy quality assurance procedure for medical physicists.

treatment outcomes. 1-5 As a complex procedure involving framebased or frameless immobilization, SRS/SBRT delivers small beams in multiple noncoplanar arcs to achieve a highly conformal dose distribution on the targets, while minimizing the dose to the surrounding normal tissue. Among many advanced techniques, the CyberKnife robotic radiosurgery system (Accuray Inc., Sunnyvale, CA, USA) has been increasingly employed for the SRS and SBRT. [6][7][8] This treatment system consists of a 6 MV flattening-filter free linear accelerator mounted on an industrial frameless robotic arm (Kuka, Augsburg, Germany). It is capable of delivering precise ablative radiation dose to the target by utilizing a large number of noncoplanar beams while simultaneously tracking target motion in real time.
After field service engineers (FSE) perform a full set of beam path calibration for all collimators, isocentric end-to-end (E2E) tests are performed to determine the overall targeting accuracy and coordinate coincidence of the CyberKnife system for each tracking method. In order to minimize the overall targeting error, a targeting correction value known as "DeltaMan" is introduced based on the E2E results to change the offset between the machine center of robot frame and the imaging center of target localization system (TLS) frame. Targeting accuracy is defined as the offset between centroid of the delivered 70% isodose line and the known centroid position in patient reference frame. The targeting error tolerance is 0.95 mm for all tracking methods. E2E tests are conducted for evaluating the performance of the overall system and the final results are influenced by both beam path accuracy and TLS. While the methodology for QA of TLS has been well established, simple and convenient QA methods for beam path accuracy still remain to be developed.
AAPM Task Group 135 recommends three levels of QA evaluations for the current state of manipulator-pointing accuracy. 9 The first level is either laser alignment check on the floor and/or automatic quality assurance (AQA) test, which is a simplified Winston-Lutz test consisting of only two beams. The second level is to run a "BB test" (an isocentric plan) in simulation mode for visually checking whether the centerline laser fully illuminates the isocrystal tip. The third level is a quantitative evaluation of the second level test that is capable of recording node-by-node deviations. Both level 2 and level 3 tests are strongly based on the position of the beam central axis laser and level three can only be done with the assistance of a FSE.
Because there is no alternative QA method for checking the pointing accuracy of individual node quantitatively, the Task Group recommended the development of a QA procedure that could be conducted easily and safely by a Qualified Medical Physicist.
A commercially available three-dimensional (3D) cylindrical diode array ArcCHECK (SunNuclear Corp., Melbourne, FL) has been shown to be a useful tool for QA of Intensity Modulated Radiation Therapy (IMRT), Volumetric Modulated Arc therapy (VMAT) and Tomotherapy treatments. [10][11][12][13] The ArcCHECK consists of 1386 diode detectors which are arranged in a spiral pattern with length of 21 cm and diameter of 21 cm. In addition to its application in dosimetric verification, its unique 3D design also makes it a potential tool for machine QA. [14][15][16] Although the ArcCHECK has been widely used for QA of isocentric treatment machines, its applications in commissioning of Monte Carlo algorithm and patient-specific QA for nonisocentric treatment machine such as CyberKnife were also studied by some groups. 17,18 In this study, we developed a novel method for testing the constancy of beam path accuracy of our CyberKnife (model M6) system by using ArcCHECK. Though it is not intuitive for a detector array with a spatial resolution of 1 cm to detect a submillimeter position shift, we should emphasis that it is the single diode with high signal drawing our attention in this study. The position error was calculated based on the correlation curves and measured dose difference of a single diode instead of the conventional profile measured by multiple diodes.

2.B | Density override and QA plans
To avoid the dosimetric uncertainty arising from the artifacts of kV CT images, the acrylic body of ArcCHECK was contoured and assigned a fixed relative electron density (RED). Since MultiPlan (CyberKnife treatment planning system) forced the RED values of air to be 0 for CT numbers between 0 and 199 inclusively, 19 we also fixed the RED of the Styrofoam. A simple two-field plan, which is similar to the conventional AQA plan, was created using only the To evaluate the constancy of beam path accuracy, an isocentric plan was created with a fixed 5 mm treatment cone. Total 116 beams in body path were selected, and 50 MU was assigned for each beam. Figure 2 shows the two QA plans generated by the Multiplan. In addition, a simple anterior beam with 200 MU was delivered every time before the delivery of QA plan to make sure the output constancy of CyberKnife was better than 0.5%. The accuracy of dose calibration of the ArcCHECK was verified by a routine check using a conventional LINAC with a 10 × 10 field size.

2.C | Correlation curves of diodes
To make sure the number of tracked beams coincided with the number of selected diodes, a complete delivery of the test plan using Arc-CHECK was first saved as a movie file (.acm file) and the screen of console computer showing the beam ID was simultaneously recorded using a video camera. When analyzing the movie file, we set several criteria to make sure that each selected diode corresponded to one beam: (a) if multiple diodes succeeded to measure the signal from one beam, only the diode with the maximum signal would be selected; (b) if one diode picked up multiple signals from different beams, this diode would not be selected. Therefore, after the movie file was synchronized with the video, the connections between selected diodes and specific beams would be confirmed. In addition, to avoid those diodes too close to the field edge, a threshold of 10% was applied, which implied that only those diodes with measured signal larger than 10% of the maximum could be selected in the analysis.
Finally a Look-Up- Table (LUT) containing a total of 68 selected diodes was saved and would be used for all future measurements.
The same CT images of ArcCHECK were anonymized and saved as a QA template in MultiPlan. Based on the isocentric plan mentioned above, we created a QA plan and then intentionally intro- finally obtained for each selected diode.
As indicated above, the most time-consuming part of our proposed method was the export and processing of DICOM dose files.
Depending on the specifications of computers, 3-4 h may be needed to obtain all necessary files.

2.D | Calculation of position error
The isocentric plan was delivered six times within 1 month and a baseline was then set by averaging these measurements. By comparing with the baseline, a map of relative percentage difference could be obtained for each measurement thereafter. Based on the dose difference map and correlation curves, position errors could be calculated in several steps while its rationality would be discussed later: Step the 'true' position error. More elaborations could be found in Section 4.
Step 2: In order to reduce the influence from the placement inaccuracy of each diode, the isocentric plans was delivered five times consecutively on the same day. Based on the assumption that CyberKnife has good reproducibility for continuous deliveries without changing the setup, a map of standard deviation (SD) over averaged measurement was generated and the position errors were then calculated based on the correlation curves for different placements of diodes (as mentioned in Section 2.C). Another LUT recording the placement of each diode corresponding to its minimum position error was saved and would also be applied for all future measurements.
Though the LUT might not represent the real position of diodes inside the ArcCHECK phantom, this step still provided a fast and easy way to minimize the inaccuracy. Step

| 113
Once the two LUTs were generated, the analysis of measurement could be streamlined. MATLAB programs were developed to process these measured dose files (.txt files) by ArcCHECK. There was no special tool box or advanced programming skills involved in this study. Analysis results could be obtained in seconds after a 20min treatment delivery.
Consecutive measurements were conducted in 6 months to test the performance of our proposed method. Since the routine AQA and E2E tests gave good results, we had no chance to detect the beam path inaccuracy in real situation. Therefore, we introduced shifts of alignment center ranging from 0.3 to 2.0 mm by pulling the Styrofoam insert out in SI direction carefully. To minimize the additional uncertainty from the TLS, we made sure the only change of robot correction after pulling out the insert was the translational shift in SI direction.

2.E | Verification of correlation curves
Measurements were also conducted to verify the correlation curves of the diodes. During the delivery of the QA plan, we manually stopped the delivery at two selected beam nodes where the response of the diodes was relatively large. Then the position shifts of robot mastering in AP, LR, and SI directions were introduced using teach pendent. The same MUs were delivered at each shift and the measured signals from the corresponding diodes were recorded.

3.B | Verification of correlation curves
As illustrated in Fig. 5, the measured correlation curves agree well with the calculated curves, especially within the range of ±1 mm.
Though it was not shown in this manuscript, the maximum percent difference between the calculated beam profile by MultiPlan and measured beam profile for 5 mm cone at 15 and 50 mm depth was <0.3%, which verified a very small uncertainty coming from the TPS modeling.

3.C | Reproducibility
As mentioned in step 2 of Section 2.D, the calculated position errors actually presented the reproducibility of robot mastering. Table 1 shows the root mean square (RMS) and the maximum of those cal- such finding and further study will be necessary.  respectively. According to Table 2, it is possible for our method to detect a systematic position error of less than 0.3 mm. Because the number of selected beam is less than the total number of beams, our method may be more sensitive to the position error. Therefore, we think the tolerance of 0.5 mm for RMS is still applicable.

| DISCUSSION
As shown in Fig. 8, it is inappropriate to directly compare the measurement with the dose distribution calculated by MultiPlan.
Not only some beams were partially or entirely missed because of the 10 mm detector spacing of the ArcCHECK, but the angular and dose rate dependence of the ArcCHECK made the direct comparison difficult. Dose to agreement (DTA) analysis was also conducted for measurements taken on different days.
As indicated in Fig. 9, the local dose difference had to be increased to~15% to achieve a passing rate higher than 90%, Though the isocentric plan used a full body path with a total of 116 beams, our study only included 68 diodes/beams. It was expected because some beams would be totally missed due to the limited special resolution of ArcCHECK. In addition, we also had to filter out those diodes irradiated by opposing or closly opposing beams, which would cancel out or exaggerate the calculated error. It is obviously that our current method is not able to evaluate the full beam path, and a true 3D dosimeter, for example, gel dosimetry system, should be more suitable for studying the targeting accuracy.
However, this study was a proof of concept about how to utilize the unique 3D structure of a widely-used dosimetry QA tool for machine QA of CyberKnife.

| CONCLUSION
This study has proposed a novel and convenient QA method for evaluating the constancy of beam path accuracy of CyberKnife, which also makes the assessment of position accuracy for single beam possible. Compared with the recommended method by the manufacturer, our proposed method is efficient and suitable for routine QA with high sensitivity.

CONFLI CT OF INTEREST
The authors declare no conflict of interest. F I G . 9. Plot of dose to agreement passing rate as a function of local dose difference for data measured on two different days.