Hi Mohamad,
It would make some sense to do that within regions that are expected to have constant intensity, the FA is not expected to be constant across WM.
The easiest way to do an SNR calculation is choose a WM area on the b=0 image and compute the mean intensity. Then identify a background region outside the brain, that does not contain any artefacts (e.g ringing), and compute the standard deviation (or the mean). Then you can divide the two and you could use some correction factors to account for the fact that the noise distribution is not Gaussian. See for details the Appendix of
http://www.ncbi.nlm.nih.gov/pubmed/17622966
A less biased way to obtain an SNR estimate when the noise is spatially variant (which is the case with the modern sequences) is to use two repeats of the same image (e.g. two b0s) and compute the mean of the sum and standard deviation of the difference within a ROI.
Cheers
Stam
On 2 Jul 2014, at 03:45, M. alshikho <[log in to unmask]> wrote:
> Hi FSL experts,
> In order to calculate the signal to noise ratio (SNR) in the right hemispheric white matter I did the following :
> 1. I used fslstats to calculate the mean and standard deviation (SD) for the right hemispheric white matter fractional anisotropy
> 2. I did calculate the SNR by dividing the mean and standard deviation SNR=mean/SD
>
> Kindly, Is this correct?
> Thanks for any suggestions or comments
>
> Mohamad
|