MRI quality control for low‐field MR‐IGRT systems: Lessons learned

Abstract Purpose To present lessons learned from magnetic resonance imaging (MRI) quality control (QC) tests for low‐field MRI‐guided radiation therapy (MR‐IGRT) systems. Methods MRI QC programs were established for low‐field MRI‐60Co and MRI‐Linac systems. A retrospective analysis of MRI subsystem performance covered system commissioning, operations, maintenance, and quality control. Performance issues were classified into three groups: (a) Image noise and artifact; (b) Magnetic field homogeneity and linearity; and (c) System reliability and stability. Results Image noise and artifacts were attributed to room noise sources, unsatisfactory system cabling, and broken RF receiver coils. Gantry angle‐dependent magnetic field inhomogeneities were more prominent on the MRI‐Linac due to the high volume of steel shielding in the gantry. B0 inhomogeneities measured in a 24‐cm spherical phantom were <5 ppm for both MR‐IGRT systems after using MRI gradient offset (MRI‐GO) compensation on the MRI‐Linac. However, significant signal dephasing occurred on the MRI‐Linac while the gantry was rotating. Spatial integrity measurements were sensitive to gradient calibration and vulnerable to shimming. The most common causes of MR‐IGRT system interruptions were software disconnects between the MRI and radiation therapy delivery subsystems caused by patient table, gantry, and multi‐leaf collimator (MLC) faults. The standard deviation (SD) of the receiver coil signal‐to‐noise ratio was 1.83 for the MRI‐60Co and 1.53 for the MRI‐Linac. The SD of the deviation from the mean for the Larmor frequency was 1.41 ppm for the MRI‐60Co and 1.54 ppm for the MRI‐Linac. The SD of the deviation from the mean for the transmitter reference amplitude was 0.90% for the MRI‐60Co and 1.68% for the MRI‐Linac. High SDs in image stability data corresponded to reports of spike noise. Conclusions There are significant technological challenges associated with implementing and maintaining MR‐IGRT systems. Most of the performance issues were identified and resolved during commissioning.


| INTRODUCTION
IIn 2014, the first patient was treated with ViewRay's MRIdian integrated 60 Co 0.35 T magnetic resonance imaging (MRI) guided radiotherapy (MR-IGRT) system. 1 Since 2017, commercial MRI linear accelerators (MRI-Linacs) with magnetic fields of 0.35 T (View-Ray MRIdian) and 1.5 T (Elekta Unity) have been treating patients. 2,3 Quality assurance (QA) and quality control (QC) guidelines for MRI are addressed by the American College of Radiology (ACR), 4 the American Association of Physicists in Medicine (AAPM), 5

and the
National Electrical Manufacturers Association (NEMA) standards. 6 Separate QA guidelines are available for conventional Linacs. 7 AAPM Task Group 117 is tasked with developing MRI QC guidelines for treatment planning and stereotactic radiation therapy (RT). QC results for MR-IGRT were reported for the ViewRay 0.35 T MRI-60 Co [Ref. 8 ] and MRI-Linac, 3 and the 1.5 T Elekta Unity. 9 The quality of the MRI was previously reported to be satisfactory for both commercial low-field MR-IGRT systems. 8,10 However, a lot of time and work was required during the implementation and commissioning of the MRI-60 Co and MRI-Linac systems to resolve performance issues prior to clinical operations. In the process, much was learned about system deficiencies and fixes that benefitted manufacturing, installation, QC procedures, and future system development.
The purpose of this study is to present the lessons learned from commissioning, operating, and performing quality control on 0.35 T MRI-60 Co and MRI-Linac MR-IGRT systems. These lessons will be categorized herein as: (a) Image noise and artifact associated with electromagnetic interference (EMI) sources; (b) Field homogeneity and linearity and their effects on image spatial integrity; and (c) System reliability and stability issues.

| MATERIAL AND METHODS
Data were acquired on ViewRay MRI-60 Co (13.6 MHz) and 6 MV MRI-Linac (14.7 MHz) systems (Oakwood Village, OH). The MRI-60 Co has three depleted uranium-encased 60 Co heads positioned 120 0 apart around the gantry. 11 The MRI-Linac has six 227kg steel shields positioned 60 0 apart around the gantry. 3 Both models are shimmed to ≤25 ppm pk-pk over a 45-cm diameter spherical volume (DSV) at each gantry angle using five higher-order superconducting shims and passive shim trays located in the gradient assembly. The MRI-Linac also uses passive shims oriented around, and mounted to, the rotating gantry to shim the steel shields. Gradient shimming is used to reduce the field inhomogeneity to <5 ppm in a 24 cm diameter spherical phantom.

2.A | Image noise and artifact
During the commissioning of the MRI-Linac, we investigated sources of EMI using the three commercial phantoms. The effects of EMI from B 0 instabilities on signal averaging were investigated for the MRI-Linac using the large ACR phantom and in vivo with the torso phased array receiver coils (with body coil transmission).
Both MR-IGRT models currently average two images to produce a 2D cine frame. The reasons for averaging are twofold: a) The original image processing (target tracking and beam gating) pipeline could not handle a throughput >4 frames per second (fps); and b) The averaged images provide enhanced signal-to-noise ratio (SNR) vs single acquisitions. Long-term averaging acquires the k-space from one image followed by the k-space from the second image, then combines the two k-space datasets and reconstructs the averaged image.
Short-term averaging acquires a line of k-space for the first image followed by the same line of k-space for the second image, and then increments the phase-encode line to acquire the full k-space in this manner. Averaging can cause or mitigate image artifacts depending on the source of the variation (e.g., physiological motion) and the type of averaging. 12

2.B | Field homogeneity and linearity
The MRI-60 Co employs a tune-up shim mode that uses phantombased field homogeneity measurements for patient shimming for both 2D and 3D acquisitions. The gradient offsets (first-order shim terms) do not vary with gantry angle.
The MRI-Linac shimming represents two changes from the MRI-60 Co. First, a standard shim is performed for each patient prior to each 3D acquisition used in treatment planning and setup. The standard shim mode acquires a field map in the patient and calculates the first-order shim currents that will provide the optimal field homogeneity for the imaging volume. Second, a phantom-based shim adjustment that varies with gantry angle is used for the 2D cine treatment acquisitions and is named MRI gradient offset (MRI-GO).
In MRI-GO, the first-order shim currents are updated as the gantry position changes based on a lookup table of gantry angles and corresponding first-order shim current settings calculated using the 24-cm diameter spherical phantom.
Field homogeneity was measured for gantry angles varying from 0 to 150 0 on the MRI-60 Co and 0-345 0 on the MRI-Linac in 15 0 increments using the spherical phantom. Measurements were made using both the tune-up and standard shim modes. The corresponding first-order shim values were also recorded. A free induction decay (FID) was acquired with the sphere centered at isocenter (TE/TR: 0.35 ms/3 s, Flip angle: 90 0 , 4 Averages, 5 Hz/point, 256 complex points). The proton spectra were fit to a Lorentzian function using a nonlinear fit algorithm, and the full width at half maximum fits were then calculated.
The original magnetic field homogeneity specification for the Functional Test Procedure (FTP) was baseline +/−1.5 ppm for the MRI-60 Co and ≤5 ppm for the MRI-Linac using the tune-up shim mode. In general, the current field homogeneity target is ≤2 ppm for all gantry angles. Spatial integrity measurements were made using the manufacturer-provided uniformity linearity phantom and the body coil for image transmission and reception. The phantom was available in two formats: one with square holes and one with round holes. Both were used herein.
The spatial integrity tests were performed by centering the grid portion of the uniformity linearity phantom at seven positions relative to isocenter (axial orientation with z = 0, coronal orientation with y = 0, and sagittal orientations at x = −12.5, −7, 0, 7, and 12.5 cm). A proprietary software program (ViewRay, Oakwood Village, OH) was used to analyze the uniformity linearity phantom for compliance (within +/−1 mm error for ≤10 cm DSV and within +/−2 mm for diameters between 10 and 20 cm DSV). Measurements were also conducted for varying gantry angles in increments of 30 0 to assess the stability of the spatial integrity.

2.C | System reliability and stability
Common reliability issues were documented from maintenance logs of MRI subsystem failures. A large homebuilt phantom ( Fig. 15) was used to test the individual phased array coil elements every month and when a coil was suspected to be malfunctioning. Sixteen 6-cm diameter holes, forming a 4x4 grid, were cut into a 61 cm × 61 cm × System stability was assessed based on monthly measurements of the Larmor frequency, RF reference amplitude, SNR of the torso coils, and image stability. The SNR was calculated using the two-image difference method and a region of interest (ROI) that covered 75% of the area in a homogeneous slice of the large ACR phantom (Slice 7 of 11 from the ACR QC prescription). 4 The SNR was calculated using the mean signal in the ROI (<Signal>) from the first image and the standard deviation (SD) of the noise in the difference image (σ Noise ):

3.A | Image noise and artifact
Sources of EMI that affected MRI quality are summarized in Table 1.

Examples of EMI in MRI are shown in Figs. 1, 2. A comparison of
short-term and long-term averaging for the MRI-Linac is shown in    Figure 11 shows an example of spatial integrity F I G . 4. EMI-related (moving metal) dephasing artifacts that occurred during gantry angle rotation from 300 0 to 320 0 in 76-year-old female patient receiving adaptive MR-IGRT for pancreatic cancer on the MRI-Linac. The numbers represent the time in seconds corresponding to each frame during the gantry rotation. The cine images used 2D sagittal TrueFISP cine (TE/TR: 0.91/2.10 ms, 60 0 , GRAPPA 2, 3.5 × 3.5 × 7 mm, 1351 Hz/pixel, 2 averages, 4 frames/s). Radiation delivery is paused during gantry rotation. Therefore, there is no degradation in treatment accuracy.

3.B | Field homogeneity and linearity
not meeting the specification due to improper gradient calibration. Figures 12,13 show the dependence of spatial integrity errors on gantry angle for the MRI-60 Co and MRI-Linac, respectively.

3.C | System reliability and stability
Common sources of past MRI subsystem failures are summarized in Table 2

4.A | Image Noise and Artifact
EMI is a key consideration for the MR-IGRT since the system combines a source of EMI (the radiation therapy subsystem) with an MRI that is highly vulnerable to EMI. The Linac poses a larger threat than the 60 Co heads to the quality of the MRI since the Linac uses a high-voltage linear accelerator and radiofrequency source to accelerate electrons. In turn, the Linac components and the electron beam are vulnerable to the magnetic fields generated by the MRI. The MRI-Linac employs both magnetic and RF shields to minimize the interaction between the Linac and MRI.
Past sources of radiofrequency interference (RFI) discovered inside the magnet room included a patient camera and a switching DC power supply for the Primalert 10 radiation monitor (Fluke Biomedical, Solon, OH). Sources of RF noise in the gantry cabling were easier to detect using the body coil because of the higher flux between the RF source and the body coil surface area. Use of phased array coils may be less sensitive to RFI since the sensitivity depends on the orientation of the coil surfaces to the noise source.
The MRI subsystem passed the NEMA SNR test specification (≥12 with body coil) despite the conspicuous RFI in the MRIs. 14 Therefore, it is critical to identify and resolve EMI sources before accepting the system based on the vendor's specifications.  For the MRI-Linac, the large volume of steel shielding in the gantry produced significant dephasing artifacts when the gantry was in motion. 16 The vendor currently pauses the real-time display of the cines during gantry rotation although the images can be observed from the MRI subsystem. However, resolution of the dephasing artifacts is desirable because there are several applications that can be applied to the real-time cines that can benefit treatment including visual respiratory feedback and motion prediction. Dynamic shimming and eddy current methods are now available on commercial MRIs that may be adapted to minimizing gantry motion-related artifacts. 17,18 The image artifacts associated with long-term averaging on the MRI-Linac indicated that there is a short-term B 0 instability and its severity depends on the gantry angle.

4.B | Field Homogeneity and linearity
The main disadvantages of MR-IGRT vs x ray based IGRT are the spatial distortions that occur primarily due to gradient nonlinearities, and secondarily due to magnetic field inhomogeneities. Distortion correction, high receiver bandwidths, and use of spin echo sequences can mitigate these distortions particularly for 3D acquisi- MRI-GO was designed to address the field homogeneity challenges of the MRI-Linac during 2D cine acquisitions and the data indicates significant improvements in field homogeneity (Fig. 7).
The disadvantage of MRI-GO was the frequent software disconnects related to the real-time transmission and processing of the gantry angle and shim currents. A recent software update for MRI-GO has reduced the impact of the software disconnects. MRI-GO and tune-up calibrations should also be verified or updated when there are changes to the system that can affect shimming (e.g., gradient driver replacement or recalibration, and main field ramp or shimming). Our spatial integrity error means and standard deviations were consistent with reported values. Our experiments indicate that the gantry angle had little effect on spatial integrity (Figs. 12, 13). This was expected since the spatial integrity is primarily dictated by the gradient linearity unless the local B 0 inhomogeneity is comparable to the pixel bandwidth. 23 It is important to rerun system tests after a major component is replaced or repaired to verify system performance. In addition, the medical physicists must be aware of the system changes conducted by the service engineers since these changes can also impact system performance. For example, after a failure of spatial integrity tests, we subsequently discovered that the vendor had incorrectly recalibrated the gradients on the MRI-Linac ( Fig. 11).

4.C | System reliability and stability
Commercial MRI-IGRT systems combine two distinct subsystems (MRI and radiation therapy delivery). In the case of the MRIdian systems, the radiation therapy control (RTC) is the master and the MRI is the slave. Communications issues or system faults from either subsystem often cause a software disconnection between the two subsystems that halts operations.
The torso phased array receiver coils are the component that  We developed a coil QC method that checks for bad coil elements using a home-built phantom.  24 We measured SNR to be 6% higher on the MRI-Linac.
The Larmor frequency varied by less than +/−3 ppm (≤2 σ) in both models over the long term. According to AAPM Report No. 10, the drift rate for superconducting magnets should be ≤0.25 ppm/day during routine operations. 5 The vendor does not have a long-term stability specification but does have a short-term stability specification of <3 ppm/hr that was met during annual QC measurements. F I G . 1 6 . Torso coil SNR measured monthly using the ACR phantom and T 1 weighted MRIs (TE/TR: 20/500 ms, 90 0 , 1 × 1 × 5 mm, 78 Hz/pixel, 260 s). ACR; American College of Radiology; MRI, magnetic resonance imaging; SNR, signal-to-noise ratio.