Automated analysis of ACR phantom data as an adjunct to a regular MR quality assurance program

Results Figures 2 and 3 show box-whisker plots of measures obtained by analyzing the phantom images from five MR scanners over a period of four months. The boxes in the plots show the range between the lower and upper quartiles of the distributions for measures and the red line shows the median. The whiskers (black bars) are placed at the extreme upper and lower points of the data that fall within 1.5 times the inter-quartile range. Outliers are plotted with red crosses. Slice thickness measure: Figure 2 shows a plot of the slice thickness estimates obtained by measuring the length of two horizontal bars in the first slice image. The bars are produced by sample in two grooves in the phantom that are precisely positioned so that the combined length of the bars is directly proportional to the slice thickness. According to ACR guidelines, the values should be within 0.7 mm of the true slice thickness, which is 5mm. Black dotted lines in the plot show the range of acceptable values. No measurements for any of the systems fall outside the acceptable range. Slice position accuracy: Figure 3 shows a plot of the slice position errors obtained by measuring the difference between the lengths of two vertical black bars in the first slice image. The bars are produced by wedge-shaped pieces of Plexiglas that, if the slice is positioned correctly, create voids of equal length in the image. The difference in lengths of the bars is directly proportional to the error in positioning of the slice. For four of the MR systems, all measures are well within ACR specifications (<2.5mm). However, some measurements on system MR5 are outside the acceptable range. In this case it was determined that the table positioning mechanism was faulty and corrective action was taken. Validation: The measures obtained by automatic analysis were validated by comparing them with those obtained by manual measurement on images using analysis tools in the OsiriX dicom viewer. As an example, Fig.4 shows a comparison of the manual and automated measurements for the bar length used to determine slice positioning error from the images on one of the five scanners. Correlation between the two measures in this case is better than 0.96. For certain measurements, such as those used for determining slice thickness, the agreement is not as close. This would be as expected because the manual measurement method is more subjective and is therefore prone to operator-based variation. Discussion The analysis reported in this work was limited to images using a T1-weighted spin-echo sequence to acquire data using the ACR phantom placed in the standard quadrature head coil. The software package can, however, be extended to analysis of images using other coils with appropriate phantoms and such a regular (e.g. monthly) quality check could be useful to identify other potential system problems. Currently all images are transferred to a central server where they are analyzed using the Matlab functions. Alternatively, the software package could be compiled and installed to run directly on the MR scanners with results logged or transmitted automatically for review by MR physicists. References: