Hello,
I notice an odd thing in SPM , that I don't realize if it's a feature or a
bug.
My understanding is that SPM5 exploits the feature of Nifti format that
allows to include an affine registration in the file header. However, it
seems that there are results unexpectedly different if the affine
registration is applied (by reslicing the images)or is only estimated,
especially if the original planes are different.
For example, consider these two simple pipelines, where only step a)
differs, and the images are ANAT:high resolution, pseudosagittal anatomic
image, EPI: low resolution, pseudoaxial EPI
1)
a) coregister (extimate) ANAT to EPI,
b) segment ANAT,
c) apply the normalization parameters obtained at b) to ANAT and obtain a
normalized version of the same.
2)
a) coregister (extimate & reslice) ANAT to EPI,
b) segment ANAT,
c) apply the normalization parameters obtained at b) to ANAT and obtain a
normalized version of the same.
The final normalized anatomical images ('wANAT' and 'wrANAT', respectively)
differ. Of course the image obtained with the pipeline 2) is smoother,
given the two resamplings it underwent, however it also fits more closely
the normalization template, being the image obtained from the non-resliced
anatomic a little out of register.
My questions are: is it possible that there is some bug that prevents the
routine to manage correctly the affine coregistration included in the nifti
header, or perhaps the better normalizazion for the approach 2 is a side
(positive) effect of the increased smoothness of the image to be segmented?
Is there an "optimal" way to manage such as case, without messing the hard
drive with tons of redundant resliced images?
Finally, more important, are there other steps where don't reslicing can
cause unexpected or bad results?
TIA & Best regards
Federico Giove
|