Quantitative analysis of image quality for acceptance and commissioning of an MRI simulator with a semiautomatic method

Abstract Magnetic Resonance Imaging (MRI) simulation differs from diagnostic MRI in purpose, technical requirements, and implementation. We propose a semiautomatic method for image acceptance and commissioning for the scanner, the radiofrequency (RF) coils, and pulse sequences for an MRI simulator. The ACR MRI accreditation large phantom was used for image quality analysis with seven parameters. Standard ACR sequences with a split head coil were adopted to examine the scanner's basic performance. The performance of simulation RF coils were measured and compared using the standard sequence with different clinical diagnostic coils. We used simulation sequences with simulation coils to test the quality of image and advanced performance of the scanner. Codes and procedures were developed for semiautomatic image quality analysis. When using standard ACR sequences with a split head coil, image quality passed all ACR recommended criteria. The image intensity uniformity with a simulation RF coil decreased about 34% compared with the eight‐channel diagnostic head coil, while the other six image quality parameters were acceptable. Those two image quality parameters could be improved to more than 85% by built‐in intensity calibration methods. In the simulation sequences test, the contrast resolution was sensitive to the FOV and matrix settings. The geometric distortion of simulation sequences such as T1‐weighted and T2‐weighted images was well‐controlled in the isocenter and 10 cm off‐center within a range of ±1% (2 mm). We developed a semiautomatic image quality analysis method for quantitative evaluation of images and commissioning of an MRI simulator. The baseline performances of simulation RF coils and pulse sequences have been established for routine QA.

tation large phantom was used for image quality analysis with seven parameters.
Standard ACR sequences with a split head coil were adopted to examine the scanner's basic performance. The performance of simulation RF coils were measured and compared using the standard sequence with different clinical diagnostic coils. We used simulation sequences with simulation coils to test the quality of image and advanced performance of the scanner. Codes and procedures were developed for semiautomatic image quality analysis. When using standard ACR sequences with a split head coil, image quality passed all ACR recommended criteria. The image intensity uniformity with a simulation RF coil decreased about 34% compared with the eight-channel diagnostic head coil, while the other six image quality parameters were acceptable. Those two image quality parameters could be improved to more than 85% by built-in intensity calibration methods. In the simulation sequences test, the contrast resolution was sensitive to the FOV and matrix settings. The geometric distortion of simulation sequences such as T1-weighted and T2-weighted images was well-controlled in the isocenter and 10 cm off-center within a range of AE1% (2 mm). We developed a semiautomatic image quality analysis method for quantitative evaluation of images and commissioning of an MRI simulator. The baseline performances of simulation RF coils and pulse sequences have been established for routine QA.

| INTRODUCTION
Compared with CT, MRI has the advantages of nonionizing radiation, superior soft-tissue contrast, and allowing quantitative or semiquantitative analysis of functional images. [1][2][3] Developments in radiotherapy require precise MRI images for target and normal tissue delineation, characterizing tumor features, and monitoring treatment response during and after radiotherapy. 4,5 MRI simulation is a relatively new technique for radiotherapy. 6,7 Because MRI simulation serves a different purpose from MRI diagnosis, the technical requirements are different. 8 To meet the needs of radiotherapy, an MRI simulator requires a scanning bore ≥70 cm, a flat couchtop, and an external laser positioning system installed in the scanner room. 9 According to AAPM Report 100, 10 the main procedures for acceptance and commissioning of a diagnostic MRI should include general system checks and MRI scanner system tests. Image quality tests play an important role in checking and monitoring the performances of an MRI scanner system. The gradient subsystem is assessed by geometric accuracy tests. Slice thickness accuracy is evaluated for combined gradient/radiofrequency (RF) subsystem.
Percent image uniformity (PIU), high-contrast spatial resolution (HCSR), low contrast detectability (LCD), and percent signal ghosting are evaluated for the performances of global system.
As of now, there is no formal technical report about the acceptance and commissioning of an MRI simulator. The image acceptance and commissioning for an MRI simulator are mostly described in AAPM Report 100, 10 and the accuracies of laser and table are dealt with in AAPM Report TG 66. 11 Several studies 12,13 have already reported and discussed the procedures and strategies of using an MRI simulator in radiation oncology department. However, overall strategies of image quality testing for acceptance and commissioning of MRI simulation still need to be explored.
The procedure of MRI simulation is complicated by comparison of CT simulation, because except the scanner, the RF coils also should be applied to receive the MR signal. 14 The parameters setting for each pulse sequence in MRI scanning are also complicated and flexible. 15 The image is dependent on a host of intrinsic parameter (the spin-lattice relaxation time, the spin-spin relaxation time, etc.) and operator-selectable parameters (repetition time (TR), echo time (TE), etc.). The image quality tests should not only reflect the hardware performance of the scanner but also reflect the features of the RF coils and pulse sequence. This may entail large quantities of image data for analysis at the MR workstation using built-in measurement tools. The whole process involved quite a number of manual operation which will be time-consuming and not easy to keep results objective enough.
MRI simulation differs from diagnostic MRI in purpose, technical requirements, and implementation. We are proposing a semiautomatic image acceptance testing and commissioning procedure for the scanner, simulation RF coils, and simulation pulse sequences of an MRI simulator. An ACR MRI accreditation phantom was used for image analysis.

2.A | An MRI simulator and testing phantom
It is a cylindrical phantom with inside length 148 mm and inside diameter 190 mm. We designed a bracket for the phantom (Fig. 1) to stabilize its position on the flat couchtop.
According to Task Group No. 66 report, the accuracies of localization laser, external laser, table movement and couchtop should be tested before image commissioning. 11 The localization lasers and external lasers were aligned with the center of the image plane using a laser alignment phantom (AQUARIUS Phantom, LAP Laser, Boynton Beach, FL, USA).

2.B | Using ACR standard sequences with a split head coil
The MRI scanner includes static magnetic field subsystem, RF subsystem, and gradient subsystem. Acceptance testing and commissioning for the quality of the images generated by the scanner is conducted using a (a) (b) standard protocol using a split head coil and standard sequences prescribed by the ACR. 16 It includes two axial spin echo standard acquisitions, a T1-weighted image (T1WI) (TE/TR:20/500 ms) and a T2-weighted image (T2WI) (TE/TR:80/2000 ms), with the settings FOV = 25 cm, thickness = 5 cm, gap = 5 cm, NEX = 1, matrix = 256 9 256. The scan direction is from chin to nose as labeled on the phantom, and the 11 standard scan layers for analyzing are labeled S1-S11.  , and T2 FSE were tested using the ACR phantom. The slice thickness is often set at 2-3 mm for simulation of head and neck cancer or brain cancer cases. For better slice localization in the ACR phantom, the setting of 2.5 mm slice thickness with no spacing was adopted. FOV was set 25 cm, and frequency-encoding direction was set at anterior/posterior (A/P). The other scanning parameters of simulation head sequences for commissioning are shown in Table 1.
For simulation body sequences, T1 in-phase images of LAVA-Flex (Liver Acquisition with Volume Acceleration with Flex processing), T1 FSE, T2 PROPELLER, T2 FSE were included in the study with FOV = 42 cm, slice thickness = 5 mm, and gap = 0, and frequencyencoding direction was set at right and left (R/L). The other scanning parameters are shown in Table 2. For testing the whole FOV for geometric accuracy, the ACR phantom was installed in both the isocenter and 10 cm off-center.

2.E | Image analysis
Seven quantitative image parameters were tested and recorded for MR simulation image acceptance and commissioning.
For geometric accuracy, the diameters of four radial lines (0°, 90°, AE45°) on S5 and S1 were auto-measured. The percent geometric distortion (%GD) was calculated separately according to the following equation: The slice position accuracy tested on S1 and S11 was auto-calculated with the difference of left and right bars separately (Eq. 2).
Half of DSP was the actual slice displacement error. When DSP >0, the slice mispositions superiorly, and DSP <0 means the slice mispositions inferiorly.
Slice thickness (ST) in MRI is ideally determined by the bandwidth of the RF excitation pulse and the amplitude of the associated applied The main scanning parameters for simulation head sequences for the commissioning protocol.

Sequences
The percent integral uniformity (PIU) for image intensity was calculated according to Eq. 3, below. A 1 cm 2 circular region of mean maximum ð S max Þ and minimum ð S min Þ gray values within the center region of a 200 cm 2 circle on S5 was automatically delineated and recorded.
For testing percent signal ghosting, four rectangular regions of 10 cm 2 were delineated for extracting mean signals in the frequency-encoding direction ( S FE1 and S FE2 ) and in the phase-encoding direction ( S PE1 and S PE2 ). The mean signal ( S) of the 200 cm 2 circle within the center region was also recorded. The ghosting ratio was calculated as: HCSR in frequency-encoding and phase-encoding directions was semiauto tested on S1. An experienced medical physicist identified three pairs of arrays of holes with resolutions of 1.1, 1.0, and 0.9 mm, respectively.
LCD was semiauto tested from S8 to S11, which represented According to the AAPM 100 report, all the image parameters using ACR standard sequences with a split head coil pass the threshold of acceptance criteria (Table 3 and Fig. 3). Figure 4 shows the automatically calculated image parameters for ACR T1WI sequence.

3.B | Testing simulation RF coils for radiotherapy with ACR standard sequences
The performance of the six-channel simulation head coil, eight-channel diagnostic head coil, and the simulation body arrays were tested by conducting axial ACR standard sequences.
For ACR T1WI and T2WI sequences, compared with the eightchannel diagnostic head coil, the PIU of the simulation head coil decreased by 34.37% and 34.04%, respectively ( controlled <2.5%. On S1 and S5, the average %GD was controlled in the range of AE 1%. Slice position could be controlled within AE0.4 cm on Slice 1 and Slice 11 by using all coils. The ST for each coil also could be controlled in the normal range. 3.C | Testing simulation pulse sequences with simulation RF coils for radiotherapy Table 6 shows the summary of image quality parameters of T1WI and T2WI clinical pulse sequences by using the six-channel simulation head coil. With regard to PIU, the values of all the sequence were <40%, and the mean value with standard deviation was 36.68 AE 0.57%, which was similar to the results of the ACR standard sequences using the same coil. For HCSR, the values in some of the clinical sequences are improved less than 1. The matrix in the phase-encoding direction was set less than in the frequency-encoding direction; the corresponding HCSR is 0.9-1 in the phase-encoding direction and 0.9 in the frequency-encoding direction (Fig. 6).
For the LCD, the values of T2-weighted simulation pulse sequences were lower than of T1-weighted simulation pulse sequences. The slice thickness of 3D T1 FSPGR was 35.2% higher than the true value of 2.5 mm. The scanning time of 3D T1 FSPGR was the shortest among the T1WI sequences. The parameters of slice position, geometric accuracy, and GR were in the normal range (Table 6 and Fig. 7). was controlled within AE1% (Fig. 8).
The slice position for all tested sequences also was kept in the normal range. The scan time of LAVA-Flex is the shortest among the tested clinical sequences.

| DISCUSSION
With the growing prevalence of MRI simulators in radiation oncology departments, it is imperative to monitor the stability of scanners, RF The automatically calculated image parameters for ACR T1WI with split head coil: (a) PIU and GR on S7; (b) and (c) Geometric accuracy on S1 and S5; (d) and (e) slice position accuracy on S1 and S11; (f) slice thickness accuracy on S1.  The intensity uniformity of MRI system is due to both the RF transmitting (B1-field) and RF receiving systems. The intensity uniformity of images using conventional diagnostic coil also could not meet the tolerance recommended by AAPM. That is because, the surface coil may lead to lose some image uniformity, although it is characterized by a high signal-to-noise ratio. 17,18 However, compared with the conventional diagnostic head coil, the six-channel simulation head coil produces more serious heterogeneity. Conducting simulation sequences with the simulation head coil, the PIU is still quite low. Liney et al. 12 reported that the radiation head coil was less homogeneous than the GE head and neck coil and the body phase array. The intensity correction method is recommended to be chosen when using six-channel simulation head coil.  Geometric distortion is complex and depends on many factors including system imperfection, patient anatomy, pulse sequence type, and image parameters. 19 System-dependent distortion mainly stems from gradient nonlinearities, static field inhomogeneities, and eddy currents created by the switching of field gradients and maladjustments of both the gradient offsets and the radio frequency. 20 For the purpose of radiotherapy, it is particularly important to check whether the geometrical accuracy of image is sufficient to allow precise target and OARs delineation. The assessment of geometric distortion for MR simulation images should include the whole FOV. In this study, we placed the phantom in the isocenter and 10 cm offcenter to test the GD using simulation body arrays. And GDs of two positions can be all controlled in the range of AE1% (2 mm). Some new large geometric phantoms are being developed to simulate and evaluate distortion with large FOV. 21,22 The magnitude of the distortions increases with increasing distance from the isocenter of the scanner. 23,24 Similar to our results, within a distance of 200 mm, the mean distortion in the axial plane can be controlled in an acceptable range for radiotherapy. Distortion in the sagittal and coronal planes can also be evaluated using the ACR large phantom.
The newly developed semiautomatic image analysis codes could make the results of measurement consistent, reduce the bias of human analysis, and save time. They improve the precision and efficiency of the acceptance test, and could be a useful software tool for routine QA procedures. The automatic image analysis procedure for both MRI and cone beam can help finish the uniform acceptance and constancy testing. 25,26 Fully automatic image quality assurance for an MRI simulator system should be carefully considered in the future to minimize manual intervention.

| CONCLUSION
Following the AAPM Report 100 for MRI, we developed a semiautomatic method to evaluate the basic image quality parameters for an MRI simulator. A series of image acceptance tests and commissioning for the scanner, RF coils, and pulse sequences have been constructed. The six-channel simulation head coil can provide comparable images, except for poor uniformity. The intensity correction method is recommended to be chosen when using six-channel simulation head coil. The baseline performances of simulation RF coils and pulse sequences have been established for routine QA.
These proposed procedures can be added as the part of an MRI simulator commissioning.