TomoMQA: Automated analysis program for MVCT quality assurance of helical tomotherapy

Abstract Purpose In this study, we developed a simple but useful computer program, called TomoMQA, to offer an automated quality assurance for mega‐voltage computed tomography (MVCT) images generated via helical tomotherapy. Methods TomoMQA is written in MATLAB and contains three steps for analysis: (a) open the DICOM dataset folder generated via helical tomotherapy (i.e., TomoTherapy® and Radixact™), (b) call the baseline data for the consistency test and click the “Analysis” button (or click the “Analysis” button without the baseline data and export the results as the baseline data), and (c) print an analyzed report. The overall procedure for the QA analysis included in TomoMQA is referred from the TG‐148 recommendation. Here, the tolerances for MVCT QA were implemented from TG‐148 recommended values as default; however, it can be modified by a user manually. Results To test the performance of the TomoMQA program, 15 MVCTs were prepared from five helical tomotherapy machines (1 of TomoTherapy® HD, 2 of TomoTherapy® HDA, and 2 of Radixact™) in 3 months and the QA procedures were performed using TomoMQA. From our results, the evaluation revealed that the developed program can successfully perform the MVCT QA analysis irrespective of the type of helical tomotherapy equipment. Conclusion We successfully developed a new automated analysis program for MVCT QA of a helical tomotherapy platform, called TomoMQA. The developed program will be made freely downloadable from the TomoMQA‐dedicated website.

quality of MVCT images performed by the quality assurance (QA) procedures is important; AAPM also strongly recommends monthly QA procedures to guarantee MVCT image quality. 5 For such procedures in helical machines, physicists acquire an MVCT image set from the Virtual Water™ phantom (normally called "Cheese" phantom), and check the differences with respect to the baseline image [e.g., a CT image acquired at the time of machine acceptance test procedure (ATP)].
The monthly QA items and their tolerance limits for MVCT are listed in the AAPM TG-148 report 5 and the vendor's manual, 6 but unfortunately, the reports do not provide specific methods for the analysis of MVCT QA. The unconstrained analysis method might be subjective depending on the analyzer (conventionally a medical physicist), and it may require considerable time to distinguish between right and wrong. Even if several third-party software for QA in tomotherapy are introduced in the TG-148 report for the MVCT QA analysis, 7 there is currently no automatic analysis tool for MVCT QA in helical tomotherapy.
In this study, we developed a simple but useful computer program, called TomoMQA, to offer automated analysis QA of MVCT images generated via helical tomotherapy, including not only TomoTherapy® (Accuray, Sunnyvale, USA) but also Radixact™ (Accuray, Sunnyvale, USA) which is a relatively new modality in radiation oncology. The program has been compiled within MATLAB (The Mathworks, Inc., Natick, MA) with a GUI interface, and to analyze MVCT QA, the program requires only two inputsthe DICOM image folder exported from tomotherapy and a previous result as baseline data.

2.A | Cheese Phantoms and CT acquisition
For MVCT QA, helical tomotherapy users typically utilize a cylindrical Virtual Water™ phantom (Gammex RMI, Middleton, WI), called a "Cheese" phantom, supplied by the vendor. The phantom has a diameter of 30 cm, a length of 18 cm, and several chamber holes and 20 plug holes for dose measurement and CT density tests, respectively.
Fiducial markers are embedded in the middle of each phantom. To scan the cheese phantom in the helical tomotherapy system, TG-148 and the vendor recommend setting the mode of image scanning to "fine", that is, a slice thickness of 1 mm for SRS/SBRT (or 2 mm for non-SRS/SBRT). The Hounsfield units (HU) data of the cheese phantom range from − 1024 corresponding to a density of zero to> 1000 corresponding to a density of the fiducial markers.
In general, two versions of cheese phantoms are utilized for QA of helical tomotherapy; there is no difference between them, except for body color and the number of fiducial markers embedded in the phantoms. Color dose does not affect any analysis of MVCT QA; however, fiducial markers are displayed differently, as shown in

2.B | Mechanism of TomoMQA
TomoMQA is written in MATLAB and is created as a graphical user interface (GUI) to facilitate easy utilization for users who need to examine the QA of MVCT in helical tomotherapy. Before QA analysis, TomoMQA reads an MVCT image dataset in DICOM format. In general, the DICOM dataset comprises several DICOM files that contain image data and various attributes that identify the metadata such as position, series number, image resolution, and pixel slice thickness. In addition, each DICOM file contains a two-dimensional (2D) HU matrix to represent each slice image. In TomoMQA, the images are automatically sorted based on InstanceNumber and reconstructed as a three-dimensional (3D) matrix array in memory based on PixelSpacing, SliceThickness, and HU matrix data.

2.B.1 | Consistency test for geometric distortions
To test the geometric distortions of the MVCT, the dimensions in the x-and z-directions (i.e., transaxial plane) should be measured from the distances between the markers. In addition, the longitudinal F I G . 1. Two versions of Cheese phantom for (a) TomoTherapy ® HD (or HDA) and (b) Radixact™ direction (i.e., y-direction) should be measured from the distances between a marker and the surface of the cheese phantom. The distances between the measured lengths in two MVCT images (i.e., axial and coronal images) and the physical lengths of the phantom should be compared, and then the difference should be evaluated within 1 mm for SRS/SBRT (or 2 mm for non-SRS/SBRT).
TomoMQA is designed to locate a marker-embedded slice (i.e., the middle of the phantom) in the MVCT image dataset as a first step and to define the middle slice of the phantom. Subsequently, the selected image is converted to a binary image to distinguish the position of the markers clearly, and the centroid of each fiducial marker is determined by using the regionprops function implemented in MATLAB. 8 Finally, the distances for the x-and z-directions in images are calculated between the markers, and the longitudinal distance is also calculated between one of the markers and a boundary of the phantom. The calculated results are compared to the baseline data.

2.B.2 | Consistency test for uniformity and noise
When assessing image uniformity, the average HU in a small region of interest (ROI) (e.g., a circle of approximately 10-mm diameter) located at the center of a specific image should be calculated and compared to ROIs located in the periphery. Subsequently, the largest difference between the HUs of the center and periphery ROIs should be calculated. If an MVCT image is used for dose calculation, TG-148 recommends that the difference should be < 25 HU. When assessing the image noise, the standard deviations (σCT) of the HUs in the central ROIs should be calculated. TG-148 mentioned that the noise levels (i.e., one standard deviation) are typically around 50-70 HU; however, the detailed information, such as an area of each ROIs, is not described in TG-148.
TomoMQA is designed such that it can select the pertinent image slice that contains a uniform section to assess the uniformity and noise of MVCT images. Typically, the uniform slice of the cheese phantom is located between the middle of the phantom and the edge of the A1SL chamber holes (approximately 25 mm thickness); hence, in here, the TomoMQA selects the slice 10 mm from the middle of the phantom. Regarding the uniformity test, a total of five circle-type ROIs are created from a center point and four cardinal points on the selected slice for the uniformity test. Subsequently, the average HU is calculated from each ROI, and the largest difference between the central HU and other HUs is determined to assess the uniformity of MVCT images. For the noise test, a big ROI located in the center of the phantom is additionally created, and the standard deviations (σCTs) are calculated for the small and big ROIs located in the center. Figure 3 shows the schematic of ROIs used in TomoMQA for uniformity and noise tests.
Unfortunately, regarding these two tests, the vendor and TG-148 recommendations do not include several test-related parameters (e.g., distance from center to the periphery site, ROI size for noise test).
Especially, the value of tolerance limits for the uniformity test is only suggested when the MVCT image is used for dose calculation. In TomoMQA, as a default, the tolerance limit of the uniformity test for only imaging usage was set as equal to the reference for other materials used in the contrast evaluation (Section 2.B.3). For convenience, in TomoMQA, the user can control the related parameters (e.g., ROI size, distance from center to periphery ROI, and tolerance limit) on GUI.
The results of the uniformity and noise tests calculated using TomoMQA are not compared with the baseline data, because the uniformity and noise values are inherent characteristics of their own images; baseline data are printed in a QA report solely for reference.

2.B.3 | Consistency test for contrast
The test for the image contrast should be conducted by inserting various density plugs supplied by the vendor. Figure 4 shows the cheese phantom with various density plugs inserted (left) and its cor-   TomoMQA is designed to assess the contrast quality based on the pertinent slice located~50 mm from the middle slice to the edge of the cheese phantom; the distance from the middle of the phantom and the end of plug-in hole is approximately 30 mm; however, an additional~20-mm depth is required to measure the liquidtype plugs (i.e., a true water container) as shown in Fig. 5.
Typically, the plug holes are positioned at the fixed location for either type of phantom; hence, the average HU of the specific area for each plug can be calculated easily if the center point of the phantom is known. In TomoMQA, the size of ROI to calculate the average HU is considered to be same as in the uniformity test. For the analysis, each calculated HU is compared with the corresponding value in the baseline that is close to the calculated HU irrespective of a plug position, and whether its error is consistently in the tolerance limits of the user is verified. All results are automatically calculated and are displayed in a table included in TomoMQA, and comparison targets (i.e., plugs) can be selected by a user in GUI.

2.B.4 | Test for spatial resolution
The test for the spatial resolution should be performed by inserting the high-contrast resolution plug supplied by the vendor. TG-148 recommends that the minimum resolution of 1.6 mm should be visible in the reconstructed CT images. Note that total 7-type holes (i.e., 0.8, 1.0, 1.2, 1.4, 1.6, 1.8, and 2.0 mm in diameter) are on the surface of the resolution plug, and the depth of the 1.6-mm holes is 5 mm in minimum. Typically, the spatial resolution has been nominally assessed by a visual inspector.
TomoMQA is designed to assess the spatial resolution of the MVCT image based on the pertinent slice located 35 mm or 90 mm from the middle slice depending on the insert direction of the plug. In TomoMQA, the desired image with the plug hole can be identified automatically, because the spatial resolution plug generally has a maximum noise value among the inserted plug-in due to its inhomogeneous structure. If the TomoMQA could not identify the plug automatically (e.g., owing to the image artifact), a user can vary the number of the resolution plugs manually in the GUI.
In TomoMQA, the image of resolution plug is detected and displayed in a figure, and its window level is also automatically set considering the HU range of the image. The analysis with respect to the spatial resolution in TomoMQA is designed as an exception to be manually conducted by a user. In practice, in contrast with the F I G . 3. Schematic of region of interests used in TomoMQA for uniformity and noise tests 4. View of the cheese phantom (left) and its mega-voltage computed tomography image (right). The various material plugs and high-contrast resolution plug are inserted as shown in the photo, and one of holes is ejected to measure an air density previous evaluations, the resolution assessment completely depends on the eye of the inspector.

2.B.5 | Generation of Report for MVCT QA
After completing the analysis of MVCT, TomoMQA can print out the MVCT QA report in a pdf format. The report comprises only one page and includes analysis metadata such as analysis results, baseline results, grades of assessment items (i.e., pass/fail), and a screenshot of TomoMQA program.

3.B | Example QA by using TomoMQA
For instance, Table 1 summarizes the example results of TomoMQA to evaluate the specific MVCT sets acquired for the check of machine issues as follows: • QA #1: artifact issue • QA #2: after fixing artifact issue • QA #3: after replacement of the MVCT detector F I G . 7. Example of mega-voltage computed tomography quality assurance report written by TomoMQA, The report includes all results with the corresponded reference data and the screenshot of TomoMQA graphical user interface As summarized in Table 1, the noise parameter of QA #1 increased compared with other parameters owing to the artifact issue, and subsequently, the noise parameter was returned after fix-

| CONCLUSION
In the present study, we developed a simple yet useful program, called TomoMQA, which can be used to analyze MVCT images generated from helical tomotherapy such as TomoTherapy ® and Radix-act™. Our test results demonstrated that TomoMQA can successfully evaluate the quality of MVCT images, while complying with the guidelines of AAPM TG-148. We believe that TomoMQA will be useful to analyze MVCT QA images acquired from helical tomotherapy, and it will help save time compared to manual analysis of MVCT QA measurements. The developed program can be freely downloaded from the TomoMQA-dedicated website, 9,10 and the program will be updated to overcome current limitations (e.g., loading speed, reported bugs) continuously.

CONFLI CT OF INTEREST
The authors declare that they have no competing interests.