> If the goal is to normalize functionals into MNI space, while taking into
> account the subject's anatomy as well as warping due to EPI distortion in
> the functional images, would the following steps seem reasonable:
> 1. coregister one functional to structural (T1, in my case)
I would select all of your other EPI volumes as "other" here. This would move
the whole dataset in the same way as the EPI scan that was actually
coregistered - and give the next stap a bit less work to do.
> 2. realign and reslice functional images to coregistered functional image
> 3. segment the structural image
> 4. use normalize:estimate with the mean functional image (from step 2) as
> the source and EPI template and modulated spatially normalised tissue maps
> of grey and white matter (mwc1X.img and mwc2X.img) (from step 3) as
> templates to estimate warps to use to write normalize functional images
This model isn't ideal, as the distortion correction should only really need
to be done in the phase-encode direction of the EPI, and the relationship
between EPI dropout and the gradients of the distortion field would not be
handled so well. However, you may still be on to something. I would suggest
taking the Segement route - rather than the spatial normalisation route
1) Write spatially normalised GM, WM and CSF (no modulation) from the
anatomical image, specifying a bounding box that covers the whole brain.
2) Smooth these a little bit (maybe about 3mm).
3) Use the segment button, specifying the EPI as the image to segment, and
changing the priors to the smoothed GM, WM and CSF from the anatomical image.
4) Write the spatially normalised EPI using the _seg_sn.mat file generated.
It might work OK - or it might not. See also the following post: