Coreg and realign are indeed similar. In both cases, they are meant to
rigid body align different images to each other. It's the details that
are different. Coreg is using mutual information to align images, which
is necessary when looking at different image modalities. That's why
coreg is the tool you want for aligning EPI to T1 scans. Realign is (I
think) doing some kind of minimization of squared difference. Unlike
coreg, realign will only work between images of the same modality, and
it's designed to correct for motion by aligning images within a scan
session. It is essentially a quick way of doing many coregistrations,
and I suppose that, in principle, one could use the coreg tool for the
same purpose. But then one would have the trouble of setting up
hundreds of coreg batches, and one would have to manually reconstruct
the realignment parameters. Someone more informed than myself can
clarify, but I would guess that there's little if any benefit in using
the MI criteria instead of squared diff. criteria, and there's a whole
lot of extra headache involved.
In this specific case, you are using coreg to line up your EPI images
with your T1 scan. If you did not do so, any normalization parameters
derived from the T1 scan would not be valid for your EPI images.
You don't want to estimate and reslice (emphasis on reslice) to the T1
image. This will put your EPI images in the same resolution as the T1,
and there's no sense in doing that. The whole point of the coreg step
as you are using it is to set up your EPI scans for normalization, and
the estimate half of coreg is sufficient for that.
As for your histogram, it's never going to look perfect. The T1 scans
and EPI scans are taken at different resolutions and have different
contrasts, so the information will never match exactly. The important
thing to do is check to make sure that they are aligned. Just click
around the top, edges, front, etc. and make sure they line up.
Hope this helps,
Ananth Narayanan wrote:
> I would like some clarification on the preprocessing steps, particularly
> Realign, Coreg and Normalization. Thus far, I've been doing preprocessing
> blindly. The order in which I do it is: Slice timing -> Realign (not
> 'Realign & Unwarp') -> Coregister -> Normalize -> Smooth.
> I have EPI functional scans, 35 slices, and I have a T1 weighted anatomical
> scan for each subject.
> My understanding of coregistration is that it is a within subject
> registration using a rigid-body model. But so is the 'Realign' step. If
> 'Realign' does the motion correction, what is the purpose of Coreg? If coreg
> is based on a 'rigid body' model, how can the EPI scans that are not in the
> same voxel-to-voxel correspondence with the hi-res T1-weighted scans, be
> registered with the same?
> I actually tried the estimate&reslice option of coreg with the hi-res
> anatomical as the reference image, the 1st volume EPI as the source and the
> remaining volumes as other images. The joint histogram it generated was very
> bad. My question is, what am I misunderstanding? Thanks.
> - Ananth
Neil R. Chatterjee
Stanford Systems Neuroscience & Pain Lab
[log in to unmask]