nonlinear registration of T1 to MNI template - a bit too enthusiastic? #1337
Description
Apologies beforehand, there may be a specimen in the 'model zoo' already that answers this question.
I'm using a script ( elastix_adni.txt -- other extensions not allowed! it is just BASH ) to warp PET brain scans from ADNI to MNI standard space, using one T1w MR image for all PET scans of one person. The recipe is:
- register the T1 to PET (rigid)
- register the T1-in-PET-space to MNI (nonlinear)
- apply this nonlinear transform to the PET
For the majority (!) of the scans, the nonlinear part of the transform goes pretty wild, as seen on these two examples. One goes as expected, the other stretches the space right above the corpos callosum and then compresses the part of the brain above that:
top: T1 image resampled to PET space, middle: T1 resampled to MNI space, bottom: PET resampled to MNI space
top: T1 image resampled to PET space, middle: T1 resampled to MNI space, bottom: PET resampled to MNI space
I started with the more traditional approach of first transforming PET to T1 which had the same deformations for warping T1 to MNI, so I switched to this order (also PET experts wanted me to "do as little as possible to the PET" --which IMHO was not an issue). Unfortunately, the order did not make a difference.
Is there a way to "curb" the nonlinear part of the transformation T1 -> MNI? As the script shows, the parameters that I use for the B-spline bit are just the file ElastixModelZoo/master/models/default/Parameters_BSpline.txt. I assumed the parameters in there would not make the standard space mapping of a (fairly normal-looking) T1w MR image go that wild?
Other than that it does what it says on the tin :)