Regarding your second question, the SNR in the b0 image will increase with the square root of the number of repeats. Therefore, if the scanner outputs directly a mean b0, you should take that into account to get a more representative SNR for an individual b0.
In general, the SNR computation is a bit more complicated by the modern imaging techniques, e.g. parallel imaging and multichannel coils and I think this is one reason that SNR is not frequently reported. The noise distribution is not spatially homogeneous, which is a requirement for the two-ROI approach to work (i.e. compute the signal and the noise from different ROIs). Therefore, you need to estimate both noise and signal from the same ROI, using multiple repeats of the acquisition. See (Dietrich, 2007, JMRI, 375-385) and (Reeder, MRM, 2005, 748-754) for more details.
Regarding the location within which SNR should be reported, I guess it depends on the application. If you do tractography, it will make sense to estimate the SNR within WM.
the SNR calculation is even more complicated
On 23 Jun 2010, at 18:51, Jenifer Juranek wrote:
> While the issue of SNR effects on DTI metrics is discussed in the published literature, I have found relatively little information regarding calculating SNR as a screening procedure for image quality. A few authors have reported the SNR range (or cut-off) of their datasets - but not many. I'm curious about the following questions...any responses/opinions would be very helpful.
> 1) Is calculation of SNR an effective screening tool for image quality in DTI research (e.g where FA and MD values in GM and WM are the primary variables of interest)?
> 2) If calculation of SNR is a "good thing to do and report", then what is the best approach? In particular, datatype for analysis (e.g. b=0, trace, DW) and location "inside the brain" seem to be important things to consider. In terms of datatype, I have seen folks use single non-averaged b=0 for each subject, and and mathematical relation between sequential b=0 images, and some have averaged their b=0 data on the scanner, so all that can be done is to evaluate the scanner-averaged b=0; I have also seen a recommendation on this listserve to use trace images. Regarding choice of location inside the brain (for the numerator of SNR, where the denominator is the sd of a location outside the brain, but still in the image)...should one calculate SNR for each structure of interest or might it simply suffice to calculate SNR for a GM region and/or a WM region?
> The following articles have prompted these questions:
> J Mag Res Imaging (2007) 26:756-67
> Am J Neuroradiol (2007) 28:724-30