The diagnostic images and summary values below are derived in part according to Dr. Gary Glover’s stability QA protocol, published in JOURNAL OF MAGNETIC RESONANCE IMAGING 23:827–839 (2006), and in part according to Siemens QA Protocol.

Stability Sequence
Siemens’s epi_stability_multi-sPlot sequence (Voxel Size: 3.1x3.1x5.0 mm, Image Matrix: 64x64, # of slices: 9, TR: 1000ms, TE 30ms).  Each of the 9 slices is analyzed independently.

 

Mean Signal Image
The Mean Signal Image is the simple average, voxel by voxel, across the 500 images.

Temporal Fluctuation Noise Image
The time-series across the 500 images for each voxel is detrended with a 2nd-order polynomial. The Temporal Fluctuation Noise Image (SD Image) is the standard deviation of the residuals, voxel by voxel, after this detrending step.

SFNR Image
The SFNR Image is a quotient, voxel by voxel, between the Mean Signal Image and the Temporal Fluctuation Noise Image.

Static Spatial Noise Image
The Static Spatial Noise Image (Even - Odd Image) is the difference, voxel by voxel, between the sum over all the even-numbered images and the sum over all the odd-numbered images. If the images in the time-series exhibit no drift in amplitude or geometry, this image will display no structure from the phantom, and the variance in this image will be a measure of the intrinsic noise.

Mean Signal Summary Value
The Mean Signal Summary Value is the average across the 15x15 square ROI placed in the center of the phantom in the Mean Signal Image.

 

SFNR Summary Value
he SFNR Summary Value is the average across the 15x15 square ROI placed in the center of the phantom in the SFNR Image.

 

SNR Summary Value
The SNR Summary Value is (mean signal summary value)/sqrt((variance value)/500 time points).  Where, the Variance Value is the variance across the 15x15 square ROI placed in the center of the phantom in the Static Spatial Noise Image.

 

Percent Fluctuation Summary Value and Percent Signal P2P Summary Value & Percent Drift Summary Value
To compute these summary values, a time-series of the average intensity within a 15x15 square ROI centered in the image is obtained.  The data is detrended with a 2nd-order polynomial fit.

Percent Fluctuation Summary Value equals 100 * (standard deviation of the residuals after detrending)/ (mean signal intensity prior to detrending).
Percent Signal P2P Summary Value equals 100 * ((maximum signal intensity) – (minimum signal intensity)) / (mean signal intensity).
Percent Drift Summary Value equals 100 * ((maximum fit value) – (minimum fit value)) / (mean signal value).

Percent Ghost Summary Value & Brightest 10% Ghost Summary Value
To compute these summary values, a phantom mask and a ghost mask is generated from the time-series.  The phantom mask is the area obtained by fitting a slightly oversized circle around the phantom image.  The ghost mask is the area obtained by shifting the phantom mask by half the image size, either way, along the phase encoding direction and then subtracting the phantom mask.

Percent Ghost Summary Value equals 100 * (mean intensity across the ghost mask) / (mean intensity across the phantom mask).
Brightest 10% Ghost Summary Value equals 100 * (mean intensity for the brightest 10% across the ghost mask) / (mean intensity across the phantom mask).

RDC Summary Value
The RDC Summary Value may be thought of as a measure of the size of ROI at which statistical independence of the voxels is lost, and is derived directly from the Weiskoff Plot – a log-log plot of coefficient of variation (CV) and the size of an ROI.  CV is defined as the standard deviation of a time-series divided by the mean of the time-series.  If each voxel is (relatively) independent of its neighbors, then CV for an ROI should scale inversely with the square root of the number of voxels in the ROI.  Thus, for a square NxN voxel ROI, a plot of log(CV) vs. log(N) should follow a declining straight line.  In practice, as N increases, the reduction in CV plateaus and becomes independent of N.  This occurs because system instabilities result in low-spatial-frequency image correlations, so that the statistical independence of the voxels is lost.  Define radius of decorrelation (RDC) as CV(1)/CV(Nmax), where Nmax is 15.  The RDC is the intercept between the theoretical CV(N) and the extrapolation of measured CV(Nmax).