Arkansas Children’s Hospital has a well established clinical spectroscopy practice. About a year ago, one of the clinical spectroscopy exams failed. The metabolite ratios for an infant with normal imaging findings were apparently abnormal. There were no other indications of any metabolic disorder. Since I am the consulting spectroscopist for Arkansas Children’s Hospital, Dr. Glasier asked me to investigate the failure. I determined that the shims on the scanner were out of spec. A call to our field service engineer fixed the problem. However, we felt that we could have prevented the failure had there been a QC protocol in place at the time. To prevent future failures, we decided to implement a spectroscopy QC protocol. Several protocols have been reported fro research sites and I use one of those protocols for my own research. However, the protocol I use for research isn’t appropriate for the clinical setting because it requires a special phantom. Other protocols I found were too time consuming. Since we had learned the hard way that QC is as important in the clinical setting as it is in the research setting, and since none of the published protocols I found were appropriate, I devised a simple short protocol for quality control in clinical MRS. I was particularly concerned with the localization of the data and, of course, the spectral quality of the data.
The protocol I came up with is straight forward. It uses the standard T/R head coil and the manufacturer’s spectroscopy phantom. It consists of a single slice axial localizer, followed by acquisition of data from 3 voxels of 8 cm3 each located along the diagonal of the phantom. For each voxel we acquire both TE 35 ms and TE 144 ms PRESS spectra, followed by a voxel image. The total exam takes about 30 minutes.
Because we’re interested in preventing as many problems as we can, we assess the image uniformity, the quality of the voxle localization and the quality of the spectral data. The image uniformity is calculated using the American College of Radiology guidelines. Basically, the difference between the brightest and dimmest areas of the phantom is calculated. If that difference is small, the percent image uniformity, or PIU, is high. This plot shows our image uniformity since we started taking measurements. As you can see, we had some problems at the beginning, but since then, we have been able to maintain good homogeneity across the phantom. Dips such as those at time point 3 and time point 18 indicate that something is wrong and we need to call for service.
To check the voxel localization, I adjusted the window to zero and the level to half the average intensity of the voxel. Such an image is shown here. I was concerned about outer volume contamination, so I looked for signal outside the voxel and found none in any of our runs. After I adjusted the window and level, I measured the edges of the voxel. Our nominal voxel size was 20 mm per side. We averaged 20.2 +/- 0.8 mm by 18.9 +/- 0.4 mm in plane. Because the voxel thickness thru-plane is determined by a 90° pulse, which has a sharper profile than the 180° pulses used to define the voxel in-plane, we didn’t measure the thru-plane thickness. We expect the thru-plane thickness to be better than in-plane. Our in-plane results show that we are within about 1 mm of our prescribed size, which is acceptable. Furthermore, we found that collecting data from 3 voxels was unnecessary. I was originally concerned that there would be significant variations across the phantom. However, this has not been the case, so acquiring data from 1 voxel is sufficient. This drops the scan time from 30 minutes to about 15 minutes.
Here are sample spectra from the phantom. We expect them to be good, with narrow line widths, good water suppression as demonstrated by the flat baseline, and no artifacts. If we saw broad lines, poor water suppression, or artifacts, it would signal us to investigate further and probably call service. Since the peaks are well-defined we can use peak heights or areas to calculate metabolite ratios, which are what we use clinically.
The results of the past year are shown in this table. As you can see, the values are quite reproducible, with the NAA/Cr ratio being 1.50 +/-0.03 at TE 35 ms and 1.64 +/- 0.04 at TE 144 ms. The results for Cho/Cr and mI/Cr are similarly good. If we get results outside this range, again it signals to us that we have a problem and we can correct it.
In conclusion, following the implementation of this protocol, we are now able to quickly detect and correct scanner problems to ensure the quality of clinical spectroscopy exams. Thus, this protocol is very useful as a monitor of system performance. The biggest problem is scheduling. Since I discovered that 3 voxels are unnecessary, we are collecting data from 1 voxel only. This drops the exam time to about 15 minutes, but with the clinical workload, even that can be hard to schedule. Ideally we would run QC every day. Realistically, we’re doing well to run it once a week. Even at that, since we’ve implemented this protocol, we’ve had no more failed spectroscopy exams.
Quality Assessment forQuality Assessment for
Routine Clinical MRRoutine Clinical MR
Diana M. Lindquist, PhD
Assistant Professor of Radiology
Clinical indication: apparently abnormal metabolite
ratios in infant with normal imaging findings.
Quality control is as important in clinical spectroscopy
as in research
Develop a simple short protocol for quality control in
Standard head coil and
3 voxels (8 cm3
Total time: 30 minutes
Following implementation of this protocol, we
are now able to detect and correct scanner
QC data are useful as a monitor of system
Scheduling still difficult even with a 15 minute