A novel and independent method for time‐resolved gantry angle quality assurance for VMAT

Abstract Volumetric‐modulated arc therapy (VMAT) treatment delivery requires three key dynamic components; gantry rotation, dose rate modulation, and multi‐leaf collimator motion, which are all simultaneously varied during the delivery. Misalignment of the gantry angle can potentially affect clinical outcome due to the steep dose gradients and complex MLC shapes involved. It is essential to develop independent gantry angle quality assurance (QA) appropriate to VMAT that can be performed simultaneously with other key VMAT QA testing. In this work, a simple and inexpensive fully independent gantry angle measurement methodology was developed that allows quantitation of the gantry angle accuracy as a function of time. This method is based on the analysis of video footage of a “Double dot” pattern attached to the front cover of the linear accelerator that consists of red and green circles printed on A4 paper sheet. A standard mobile phone is placed on the couch to record the video footage during gantry rotation. The video file is subsequently analyzed and used to determine the gantry angle from each video frame using the relative position of the two dots. There were two types of validation tests performed including the static mode with manual gantry angle rotation and dynamic mode with three complex test plans. The accuracy was 0.26° ± 0.04° and 0.46° ± 0.31° (mean ± 1 SD) for the static and dynamic modes, respectively. This method is user friendly, cost effective, easy to setup, has high temporal resolution, and can be combined with existing time‐resolved method for QA of MLC and dose rate to form a comprehensive set of procedures for time‐resolved QA of VMAT delivery system.


| INTRODUCTION
Volumetric-modulated arc therapy (VMAT) is a modern radiation treatment technique that allows a precise three-dimensional (3D) radiation dose to be delivered as the gantry rotates through one or more arcs. [1][2][3] Compared to intensity-modulated radiation therapy (IMRT), where the gantry is static during dose delivery, VMAT offers more precise target coverage with lower monitor units (MU). 3,4 Furthermore, the delivery times for VMAT treatments are shorter compared to IMRT treatments, 5 potentially reducing the intra-fractional patient movement during deliveries. 6 VMAT deliveries require a complex treatment plan, involving three key dynamic components; (a) gantry rotation, (b) dose rate modulation, and (c) multi-leaf collimator (MLC) motion, which are all simultaneously varied during the delivery. [2][3][4][5][6][7] It is essential to develop an independent quality assurance (QA) program for these three components of VMAT deliveries. 8 Commissioning and QA procedures for VMAT can be conducted using electronic portal imaging device (EPID). One of the most commonly used QA procedures for Varian linacs (Varian Medical Systems, Palo Alto, CA, USA) are the Ling tests, which rely solely on integrated EPID images. 9 These tests, like most common QA methodologies, focus on the accuracy of the dose rate and MLC control systems rather than the gantry angle and gantry speed aspects of the VMAT delivery. A limitation of these tests is that the EPID panel rotates with the gantry and so the accuracy of the gantry angle during VMAT is not independently assessed. Furthermore, it has been shown that a slight misalignment of the gantry angle could severely affect the dose distribution of VMAT plan deliveries, which could result in serious clinical consequences due to the steep dose gradients and complex MLC shapes involved. 10,11 For this reason, it is essential that the accuracy of the gantry angle during dynamic arc deliveries is assessed on a regular basis.
Systems for independent gantry QA have been developed by a number of groups. Adamson et al. (2012) proposed the use of a custom-built phantom with five gold coils of 0.8 cm diameter embedded in Styrofoam. Gantry angle determination was performed by acquiring cine-EPID images and extracting the projection of the gold coils to determine the gantry angle. The accuracy of this technique was characterized to be 0.0°AE 0.3°for static and 0.2°AE 0.2°for dynamic gantry rotation. 12  developed the radiographic gantry-phantom and a correction method to boost the accuracy of EPID image read-out for gantry angle. Here, a boxcar time delay correction method was applied to the gantry angle from cine EPID image headers resulting in an accuracy of 0.10°AE 0.3°(mean AE 1 SD). 14 These methods are not suitable for regular QA as they are too resource and time-intensive to be applied on a regular basis. Furthermore, most methodologies rely on the purchase or construction of a custom-built phantom. A more time-efficient and easily assessable technique is required in order for radiotherapy centers to perform independent time-resolved gantry angle QA routinely.
A fast and accurate method for dynamic gantry angle measurements was developed by   11 who used a ball bearing (BB) phantom placed on the couch and cine-EPID imaging. The projection of the BB phantom onto EPID images was used to calculate the gantry angle as a function of time. Based on this investigation, the accuracy of gantry angle determination was 0.20°AE 0.16°and 0.05°AE 0.10°(mean AE 1 SD) for static and dynamic gantry rotation using BB phantom, respectively. 11 This method has potential for use as a routine QA tool with easy setup and low cost of equipment. A limitation of this technique is that it cannot be performed simultaneously with QA of other key VMAT components (i.e., dose rate and MLC) due to the presence of the BB phantom.
Time-resolved commissioning and QA of VMAT delivery systems have been demonstrated by a number of groups. [15][16][17][18] Many of these QA procedures, for example, those which rely on machine log files or information within the EPID image header, rely heavily on the gantry angle readout from the linear accelerator itself to synchronize measurements to the treatment plan. This is also the case for a number of EPID-based patient-specific QA techniques 19,20 and delivery verification systems. 21,22 Although, as the hardware is developing, the EPID image header of TrueBeam linac has been shown to be more accurate, 23,24 this alone cannot be used for gantry angle QA as it is not independent of the linac control system. The log file and EPID header gantry angle might be accurate while all is working well, but if there is a drift or fault which results in loss of gantry calibration accuracy then this will not be evident in the log file or EPID header. Therefore, it is fully independent the method presented in this study will be sensitive to such a change and hence is suitable for routine QA.
In order for these systems to be fully independent and sensitive to all types of delivery errors (including gantry angle errors), a method is required which can accurately and easily measure the dynamic gantry angle as a function of time during acquisition of EPID images or phantom measurements. In this work, we developed a simple, inexpensive, and fully independent gantry angle measurement methodology that allows quantitation of the gantry angle accuracy as a function of time. Our proposed method is inherently independent of the linear accelerator system and does not impede phantom or EPID-based QA of MLC and dose rate, which can be performed simultaneously with this measurement. The accuracy of this novel method was evaluated for both static and dynamic gantry angle plans and for test plans designed specifically for dynamic gantry angle QA. The procedures and techniques developed in the work can be used to complement the existing time-resolved MLC and dose rate QA methods for VMAT control systems.

2.A | Overview and setup
The gantry angle measurement technique is based on the analysis of video footage of a "Double Dot" pattern attached to the front cover of FUANGROD ET AL.
| 135 the linear accelerator. The pattern consists of red and green circles of different size printed on an A4 sheet of paper using a conventional printer. Placement on the linear accelerator should be approximately near the axis of rotation but exact position is not critical. A standard mobile phone camera is placed on the couch to record the video footage during gantry rotation. The video file is subsequently analyzed and used to determine the gantry angle from each video frame using the relative position of the two dots (see Fig. 1).

2.B | Gantry angle determination
Determination of the gantry angle is based on measurement of the relative angle between the red and green dots (see Fig. 1). The method used to extract this angle from each frame of the acquired video footage can be divided into three parts: dot detection, angle determination and, gantry angle calibration. An overview of the algorithm is given in Fig 2, which is discussed in greater detail in the following sections.

2.B.1 | Dot detection algorithm
Following acquisition of the video footage, the colourmap matrix (R-red, G-green, B-blue) is extracted from each frame. Image frames corresponding to the red and green channels are used to localize the two dots separately. This effectively filters out other features of the image to maximize the accurate detection of each of the dots. For each image (red and green), the image noise is reduced by applying a median filter with 5 9 5 pixels 2 window size. Second, a 50% global threshold is used to identify the outline of the circle, which is subsequently used to locate the centroid of the circle.

2.B.2 | Angle determination
This process outlined above is repeated for both the red and green images. The relative locations of the centroids are first used to determine which of the four quadrants the gantry is in, to differentiate between the actual gantry angle h and the angle h + 180°. Second, the angle between the centroid of the red and green circles is calculated using eq. 1 where R x is the x-coordinate of the red circle, R y is the y-coordinate of the red circle, G x is the x-coordinate of the green circle and, G y is the y-coordinate of the green circle. An example of the output of the system for a single frame is given in Fig. 3.

2.B.3 | Gantry angle calibration
After extracting the relative angle between the red and green circles, this angle is then converted into the actual gantry angle. To do this, image frames from the first 10 s of the video footage are used to calculate the calibration angle (h 0 ), which is the angle between the two circles when the gantry is at absolute zero. This technique enables the method to be independent of the absolute positioning of the A4 paper on the gantry. The gantry is initially set to absolute 0°using a spirit level rather than relying on the readout of the linear accelerator gantry angle. The DynaLog files are generated by the Varian MLC control software with updated information every 50 ms, from which the gantry angle can be extracted as a function of time. The OBI gantry angle encoder signal is very precise (AE0.05°) 19 and is primarily used for conebeam CT image reconstructions. This raw signal was extracted from the header of "dark field" image frames from the KV imager using an existing external frame grabber computer. 17

2.D | Validation of methodology for static gantry
To assess the accuracy of the Double Dot method for gantry angle determination, the gantry was positioned at 10°intervals from À180°to +180°using the linac readout system. At each gantry angle, while the gantry was static, the gantry was measured using the Double Dot method, OBI gantry angle encoder and digital inclinometer (NG360).

2.E | Dynamic gantry angle QA plans
Three VMAT test plans were designed for gantry angle QA; constant gantry speed, gantry speed transitions, and maximum gantry speed During delivery of these plans, gantry angle versus time data was acquired using DynaLog files, the OBI gantry angle encoder and, the Double Dot method. Note that, in dynamic-gantry mode, inclinometer measurements were not performed due to (a) the low sample rate of the inclinometer (1 Hz) and (b) the interference between the wire connection from the inclinometer and the video.   The proposed method is practical and the associated cost is very low as it only requires a small camera (phone) and paper. This method is user friendly, easy to setup, and has high temporal resolution equal to the frame rate of the camera. Here, 30 frames per second were used, but the method can also utilize any higher frame rate resulting in a correspondingly higher temporal resolution.

| DISCUSSION
With the acquisition rate, the system can easily be synchronized to EPID images using a linear interpolation method. A limitation of synchronization between EPID and the proposed system, however, still exist in identifying the first frame of beam-on status. In our system, we assumed that the first frame of gantry rotation represents beam-on. The accuracy in synchronization between the EPID images and the Double Dot system can be improved by, for example, placing an additional dose rate meter, such as Automess Dose Rate Meter-6150 AD5/6 (Automess, Ladenburg, Germany) next to the mobile phone. Once the radiation is on, the dose rate meter will alarm while video is recoding. This alarm sound can be processed to identify which frame represent "beam-on" status for synchronization.
The ability to accurately synchronize the video signal with the Dynalog or Encoder signal is difficult. Figure 7 shows that in regions of high gantry speed, a synchronization difference results in larger gantry angle differences between the Double Dot and both Dynalog and Encoder gantry angles. It is also apparent that the encoder always leads the Double Dot and the Dynalog always lags. This produces apparent gantry angle errors in opposite directions and also causes the direction of the gantry angle error to change when direction of gantry angle rotation is reversed as shown in Fig. 7. The impact of this synchronization is also seen in Fig. 6 were a gantry angle difference of 1°is observed at gantry 0°for the dynamic test cases, however, no such error is found for the static gantry measurements in Fig. 5.
In this study, the double dot with red and green colours are used, however, different patterns can also be used. We tested on a green arrow pattern and developed an algorithm to determine the gantry angle based on Hough transform algorithm. 26 The key is in the image processing method to determine the gantry angle from video. However, we found that the double dot pattern provides highly accurate gantry angle information and is based on a straightforward algorithm, which facilitates an application and implementation. Our proposed double dot pattern has the potential to be permanently attached on the machine. Note that, the size of the dots was also different so that the dot radius could be used as a  It should also be noted that our method is not limited to using a mobile camera, but any other cameras (e.g., a video camera or web camera) can be used. In this study, we proposed the mobile camera for its practicality; (a) it is easily accessible, (b) it can record highresolution video, (c) the output video file can be easily transferred to a computer. However, it is not recommended to put the mobile camera close to the radiation beam. When using Double Dot, the room light should be turned on to increase the image contrast and to reduce image noise.

| CONCLUSION
A low cost and independent gantry angle measurement tool using a mobile camera and double dot pattern was developed and its accuracy was evaluated in this paper. The advantage of our method is that allows a simultaneous independent measurement of EPID dosimetry, geometry and gantry angle in a single delivery. The Double Dot pattern is placed on the linac, and its motion due to gantry