Fundamental study on quality assurance (QA) procedures for a real‐time tumor tracking radiotherapy (RTRT) system from the viewpoint of imaging devices

Abstract Purpose The real‐time tumor tracking radiotherapy (RTRT) system requires periodic quality assurance (QA) and quality control. The goal of this study is to propose QA procedures from the viewpoint of imaging devices in the RTRT system. Methods Tracking by the RTRT system (equips two sets of colored image intensifiers (colored I.I.s) fluoroscopy units) for the moving gold‐marker (diameter 2.0 mm) in a rotating phantom were performed under various X‐ray conditions. To analyze the relationship between fluoroscopic image quality and precision of gold marker coordinate calculation, the standard deviation of the 3D coordinate (σ3D [mm]) of the gold marker, the mean of the pattern recognition score (PRS) and the standard deviation of the distance between rays (DBR) (σDBR [mm]) were evaluated. Results When tracking with speed of 10‐60 mm/s, σDBR increased, though the mean PRS did not change significantly (p>0.05). On the contrary, the mean PRS increased depending on the integral noise equivalent quanta (∫NEQ) that is an indicator of image quality calculated from the modulation transfer function (MTF) as an indicator of spatial resolution and the noise power spectrum (NPS) as an indicator of noise characteristic. Conclusion The indicators of NEQ, MTF, and NPS were useful for managing the tracking accuracy of the RTRT system. We propose observing the change of these indicators as additional QA procedures for each imaging device from the commissioning baseline.

and liver, but its clinical use has been contributing to the reduction of planning target volume margins and improvement of patient localization based on the gold marker. 3 Quality assurance (QA) and quality control are important for radiotherapy. The American Association of Physicists in Medicine (AAPM) Task Group 142 (TG-142) Report (2009) describes the recommended methods specifically. 4 The lists of accessories for radiation treatment devices given in that report include radiographic imaging. The RTRT system is one of the "planner kV imaging" devices. In that section of the report, the recommended QA procedures (geometric accuracy, image quality, fluoroscopic dose, etc.) are described clearly. In focusing on the image quality item (spatial resolution, contrast, uniformity, and noise) in the report, the baseline data from the commissioning are recommended as criteria for QA. In addition, the RTRT system also has the aspect of an X-ray device for use in diagnostic radiology. As a standard metric for this class of device, the International Electrotechnical Commission (IEC) 62220-1 Standard (2003) is now used extensively. This standard describes the methods of image quality evaluation with quantitative indicators of digital imaging devices for medical use, involving the evaluation of the modulation transfer function (MTF) as a resolution characteristic, the noise power spectrum (NPS) as a noise characteristic, and the detective quantum efficiency (DQE) as a detector performance. 5 The RTRT system has recently undergone several improvements.
SyncTraX (Shimadzu Corporation, Japan), with a colored image intensifier (II), began to be used clinically at Hokkaido University Hospital in July 2014. 6 SyncTraX was jointly developed by Hokkaido University as a general-purpose RTRT system and can be linked to a Varian Medical Systems linear accelerator. To date, independent verification of its geometric accuracy and tracking performance has been carried out within the framework of QA of this system, but the association between three-dimensional (3D) tracking accuracy and image quality has not been analyzed. In addition, this system differs markedly from ordinary imaging devices for medical use in terms of the object's geometric system and purposes of use.
The present study was performed to propose QA procedures for this system, which covering all of the AAPM TG-142 planner kV imaging items among the fluoroscopic image quality applicable to radiation treatment devices and assuring satisfactory accuracy of 3D tracking of gold markers with this system. For this purpose, we analyzed the relationship between the fluoroscopic image quality and the accuracy of the gold marker coordinate calculation using characteristic indicators for the RTRT system. the center of the X-ray axes intersecting at the isocenter. 2 The geometric parameters are also displayed in Fig. 1. The X-ray tube enables free setting of the tube voltage in the range 40-110 kV, the tube current in the range 10-200 mA, and the pulse width in the range 1-4 ms. The colored phosphor unit is made of Y 2 O 2 S:Eu. The light emitted from this unit is quantized into three components (red, green, and blue [RGB]), each consisting of 8 bits (0-255 gradations). 6 The actual field of view (FOV) of the colored II is 228.6 mm. The gold marker at the isocenter is magnified by the geometric system. Thus, the imaging area is calculated to be 123.18 mm in diameter around the isocenter. This area is defined as the effective field of F I G . 1. The prototype SyncTraX used in the present study. Device A and Device B are both composed of an X-ray tube installed under the floor and a colored II installed on the ceiling. The experimental geometry for device A is shown. view (EFOV) as opposed to the actual FOV. Because the charge-coupled devices (CCD) has a resolution of 1000 × 1000 pixels for the EFOV, the pixel size of the colored II is deemed to be 0.123 mm. With the RTRT system, the 3D coordinates of the gold marker are calculated at a rate of 30 frames/s from a pair of two-dimensional (2D) fluoroscopic images. The 2D coordinates of the gold marker on the fluoroscopic images are determined by pattern matching.
Pattern matching employs a 2D model image (usually 24 × 24 pixels) of the gold marker registered with the software in advance as a template image. The coordinates with the highest pattern recognition score (PRS) based on the normalized cross-correlation within the search area (usually 64 × 64 pixels) adopted as the 2D coordinates of the gold marker. The normalized cross-correlation formula PRS is given by Here, N is the total number of pixels, G i is the pixel values of a template image, and F i is the pixel values of a fluoroscopic image.
The PRS ranging from 0 to 100 is calculated from multiplying 100 by the square of the normalized cross-correlation and is set to 0 when the normalized cross-correlation is negative. 2,7 In this process, the size of the tracked gold marker (diameter 1.5-2.0 mm) needs to be matched to the template image. With the colored II having three components (RGB), the PRS is calculated separately for each component, and the 2D coordinates for the component having the highest PRS are used. As shown in Fig. 2, the midpoint of the common vertical line connecting these vectors is used as the 3D coordinates of the gold marker, and the length of this common vertical line is defined as the distance between rays (DBR).

2.B | Experimental procedures and tools
A rotating phantom was used for tracking accuracy verification  devices as well as changes in 3D coordinates (σ 3D ) using the following formula: Here, σ x , σ y , and σ z denote the standard deviation (SD) of the 3D coordinates (x, y, z) for 300 frames. When the object is still, zero is ideal, but in practice, 3D statistical variations are present. In this study, keeping the 3D tracking accuracy within 0.5 mm at a probability of 99% was set as the goal, with the lower limit set at 3σ 3D < 0.5 mm and the objective limit set at 3σ 3D < 0.25 mm. Based on such constraints, the relationship of 3σ 3D to the SD of DBR (σ DBR ) and the PRS on one side was evaluated for Devices A and B. However, for the colored II with which 2D coordinates were calculated based on maximum PRS of RGB, the mean and SD of maximum PRS for RGB were evaluated.
Furthermore, under the same fluoroscopy setting, the gold marker was tracked for 10 s or more at varying rotation rates of the phantom in the range of 0-60 mm/s (at intervals of 10 mm/s).
The data from a 10-s period (300 frames) were extracted in this In this case, the gold marker with motion did not allow evaluation of σ 3D , and so, the correlation of the mean PRS and the marker speed to the σ DBR under the identical fluoroscopy settings were evaluated.

2.D | Fluoroscopic image quality and PRS
We analyzed the relationship between the 2D fluoroscopic image quality and PRS so that we could evaluate one part of this system.
Whereas the 2D coordinates from two devices are necessary for calculation of the DBR, the PRS is provided in each RGB components from each device.
MTF is used as an indicator of resolution characteristic of digital imaging devices for medical use. The IEC 62220-1 recommends the edge method for the evaluation of MTF 5 ; however, we applied a method such that the ROI is set on the image taken with approximately 45°inclination of the line pair chart and the MTF is evaluated on the basis of the mean and SD of its pixel value. [8][9][10][11] Since this method enables evaluation while placing the line pair chart at the isocenter with the use of the linear accelerator couch, the procedure is optimal for the geometric system. The formula used for calculation of MTF is given below. 8 Here, σ f denotes the SD of the ROI's pixel value in the spatial The formula for calculation of NPS defined in IEC 62220-1 is shown Here  (5) is an image quality indicator encompassing elements of resolution, noise, and contrast and is expressed in spatial frequency spectrum as is the case with MTF and NPS.

3.A | Relationship between 3D tracking accuracy and PRS/DBR
First, we confirmed that the 3D tracking accuracy was within 3σ 3D < 0.25 mm (σ 3D < 0.076 mm) when we assumed 80 kV, 200 mA, and 4 ms equivalence as a reference fluoroscopic condition in both devices. Figure 5a,b shows the relationship between the 3D tracking accuracy for the still gold marker (3σ 3D ) and unilateral mean PRS and σ DBR , respectively. A negative correlation between PRS and 3σ 3D is seen in Fig. 5a, and mean PRS > 54.34 (Device A) and mean PRS > 57.29 (Device B) were needed to achieve the lower limit of tracking accuracy 3σ 3D < 0.5 mm when the object was still. To achieve the objective limit 3σ 3D < 0.25 mm, mean PRS > 80.31 (Device A) and mean PRS > 82.27 (Device B) were needed. In Fig. 5b, positive correlation is noted between σ DBR and 3σ 3D , and σ DBR < 0.45 mm (Device A) and σ DBR < 0.39 mm (Device B) were needed to achieve the lower threshold of tracking accuracy 3σ 3D < 0.5 mm when the object was still. To achieve the objective limit 3σ 3D < 0.25 mm, σ DBR < 0.15 mm (Device A) and σ DBR < 0.17 mm (Device B) were needed.     There were interindividual differences in the input/output characteristics of Devices A and B, but it was possible to get the mean pixel value close to 128 (center of the 8-bit range, 0-255), that is,  When NEQ is evaluated, NPS is often divided in advance by S 2 of formula (5) to yield NNPS. Table 3 Table 4. Figure 9 plots the relationship between ∫NEQ and mean PRS.

| DISCUSSION
When compared with the previous IEC 61267 (1994), the IEC 62220-1 has shifted to image evaluation methods using objective F I G . 7. The relationship between the gold marker speed and σ DBR with Device A and B setting of tube voltage 80 kV, tube current 200 mA (50 mA on one side), and pulse width 4 ms.
F I G . 8. The input/output characteristics of RGB components for each device.
indicators instead of subjective indicators. 5,16 In the present study, we evaluated the image quality on the basis of methods in the IEC 62220-1, though we adopted another more optimal method in several points. The input/output characteristic of this system is proportional with X-ray exposure, and direct evaluation based on fluoroscopic image pixel value was possible for MTF and NPS evaluation. In evaluation of MTF, the methods adopted in the present study have already been evaluated in published studies on QA of electric portal imaging devices, demonstrating excellent simplicity and efficiency because direct evaluation from images is possible. 11 However, the NPS defined in IEC 62220-1 contains the term for correction; we adopted it directly for this study, taking into account also that the quadratic polynomial equation was most effective. 1D NPS was evaluated with averaging of 2D NPS recommended in IEC 62220-1. 5 NEQ can be calculated from MTF and NPS of the identical spatial frequency as an image quality indicator. Therefore, we could correlate the tracking accuracy with fluoroscopic images from the RTRT system. The NEQ can be applied to other devices using digital planner kV imaging, but it is necessary to establish QA procedures for each device with characteristics and limitations.
In the results of Fig. 8, there were interindividual differences in the input/output characteristics of Devices A and B. And in the results of Fig. 9, the PRS remained small with higher NEQ in Device A blue, and the NEQ needed to exceed the limit was about 1.5-3 times in Device A. These characteristics occurred due to the interindividual variances in II including the colored phosphor unit, the CCD, and the optical lens. The prototype system in the present study was not for a commercial system, lacking detailed adjustments.
However, we have confirmed the repeatability in another study using the prototype system. In other words, we can use the indicator also for adjustment of the interindividual difference of fluoroscopic devices.
The mean PRS has a relationship with the 3D tracking accuracy under the static condition. However, in the case of images with the mean PRS values of 60 or less, the 2D FFT could not be performed due to insufficient pixel values. As a result, NNPS and ∫NEQ could not be calculated, and the relationship between the image quality and the mean PRS was investigated in the mean PRS of more than 60. Since the real-time tumor tracking system finally performs the 3D coordinate calculation using the digital value of the fluoroscopic image, it is reasonable that a strong correlation was found between the ∫NEQ and the 3D tracking accuracy. In other words, evaluating      imaging not related to RTRT systems, a report is available concerning QA of onboard imager (OBI). 19 In that report, resolution was evaluated on the basis of visibility (highest lp/mm allowing macroscopic check). For OBI which is primarily used for the purpose of patient setup, such visibility is important and an appropriate method of evaluation. Since in the real-time tumor tracking system, pattern matching is performed based on pixel values and used for coordinate calculation, QA using objective indicators is effective in terms of the simplicity of the procedure, ease in control, and high precision.
As a summary of this study, we propose indicators for QA at two aspects of a diagnosis imaging device and a radiation treatment device in the RTRT system. However, because those indicators are affected by the interindividual difference of fluoroscopic devices, we cannot recommended the acceptable range, as indicated as the "tolerance level" or "action level" in the AAPM TG-142. 4 At some facilities, a 1.5-mm gold marker is inserted into some sites, in addition to the 2.0-mm gold marker, and it is necessary to verifying the relationship between ∫NEQ and PRS for each size of the gold markers differing in the pattern match template. 20 Some facilities use a Visicoil as a fiducial marker in a planner kV imaging. If marker templates of the Visicoil for several X-ray incident angles are generated from a breath-hold computed tomography (CT), pattern matching will be improved, and the misregistration can be reduced in kV images with sufficient image quality. 21 Some facilities use markerless tracking with a pre-acquired image using pattern matching. 22,23 In the case of pattern matching with X-ray images, pre-acquired images of multiple respiratory phases are used as multiple reference templates. 22 In the case of pattern matching with planning four-dimensional computed tomography (4DCT) images, digitally reconstructed radiographs from one phase of a planning 4DCT are used as reference template for kV images. 23 Therefore, we propose to use those QA procedures and indicators for another planner kV imaging system using pattern matching regardless of the type of marker used and with or without markers. Recently, new technologies for tracking of respiratory motion using positron emission tomography (PET) and magnetic resonance imaging (MRI) has also been reported. 24,25 However, in terms of versatility and popularity, tracking technologies using kV planner imaging will likely be the mainstream for the time being in radiotherapy. Especially in planner kV imaging systems using tracking with pattern matching, those QA indicators are well worth considering for assuring accuracy of 3D tracking.

| CONCLUSION S
In this study, image quality indicators for the fluoroscopic images of the RTRT system were evaluated with simplified procedures. The relationship between ∫NEQ and the gold marker 3D tracking accuracy was clarified through analysis of PRS. The study revealed that QA with the use of indicators, such as input/output characteristics, MTF, NPS, and ∫NEQ, was appropriate for covering all of the AAPM TG-142 planner kV imaging items, assuring the tracking accuracy of this system. In conclusion, we proposed that those image quality indicators should be added for robust QA of the RTRT system.
gave advice on the experiment part, and H.S. is a supervisor of S.K.
All authors reviewed the manuscript.

CONFLI CT OF INTEREST
No conflicts of interest.