There is no automatic way of assessing registration quality (as the best measures we have are the ones used to drive the registrations) in a way that is very meaningful across different image pairs. You can assess the magnitude of the deformation by getting a field warp output (in relative convention without the affine component) by using fnirtfileutils with the coef warp (or directly use the --fout result from fnirt) and then get the average magnitude of these deformation vectors. The latter can be easily done in fslmaths using:
fslmaths warpfield -mas brainmask -sqr -Tmean meandef
fslstats meandef -M
The fslmaths lines requires a brainmask in the reference space in order to stop it including warp vectors outside of the brain. The fslstats command takes the mean of all the non-zero vectors, which should be all the ones inside the brain as it is unlikely that any of these will be exactly zero.
If you want to be a bit more precise and allow for non-zero vectors inside the brain then you can do: "fslstats meandef -m", "fslstats brainmask -v" and "fslstats brainmask -V" and take the number from the meandef multiply it by the first number from the "brainmask -v" result (the total number of voxels) and divide by the first number from the "brainmask -V" result (the number of non-zero voxels).
All the best,
On 16 Apr 2013, at 18:13, Rali Dimitrova <[log in to unmask]> wrote:
> Apologies for this question in advance but what is the best way to compare the registration quality (or to quantify the degree of deformation applied to the FA maps) to standard space obtained using two different registration templates? Could the FA_to_target.* files be used to compute or extract the amount of warping required for each target across the FA data ?
> Thank you.
> Best wishes,