The target image is the image that is considered as remaining in its original position.
So if you wish to match a structural image to a series of (registered) functional images,
then the target would be a good representative functional image (or the mean of the series).
In SPM99, to co-register a structural image to your EPI images you would specify that the
target image was EPI. In SPM96, there is no EPI template image, so the closest template to
use would be the T2 template (but this admittedly does not always work that well).
The object image is the image that gets moved to match the target image. If you have a
structural T2 image that you are matching to your functionals, then the object would be
the T2.
The other images are a series of images that are already assumed to be in register with
the object image. The coregistration finds a transformation to match the object to the
target, then it applies the transformation to the object and also all the images that
you wish to move along with the object. If there are no other images that you wish to
piggyback with the object image, then you can just click "Done" without specifying any
images.
The alternative to moving the structural to match one of the functional images would be
to register one of the functionals with the structural, and include all of the other
functionals as "other". This way, the structural image will remain in the same position,
and the functional images will be moved (by changing their ".mat" files).
> And if I don't reslice then what have I lost? That is, is the co-registration
> calculation something that can be used later without reslicing?
If you don't reslice, you haven't really lost anything (in SPM99b). The co-registration
will modify the ".mat" files of the "object" and "other" files, to reflect the images'
new positions. The relative positions of the images (from the ".mat" files) are used
throughout SPM99b.
>
> A question I also have is this, since I have done realignment, should the
> co-registration actually calculate the co-registration parameters by comparing
> the T2 structural to one functional and then apply this calculation to the rest
> of the functional images?
You could use the structural as the "target", a mean functional as the "object"
and all the rest of the functionals as "other".
Alternatively, you could use the mean functional as "target" and the structural
as "object", and nothing as "other".
I prefer the latter, as it is generally more simple. Also, I would not suggest
doing any reslicing in the former case, because you would soon run out of disk
space as you zoom all your functional images to the same size as your high-res
structural.
>
> In a sense I would actually like to save time by having the structural T2 be
> resliced instead of the functional. To do this would I switch the roles of
> target and object? And still the question lingers what is "other" referring to?
If you specify reslicing, then the "object" and "other" images are resliced to
match the "target".
>
> Finally, does co-registration buy me anything with respect to the statistical
> analysis? Does SPM discriminate between grey, white and CSF when the
> probabilities of activation are being calculated? Or is co-registration just
> for the purpose of presentation? (I presume the same with normalization).
Co-registration is normally needed only for presenting the results on a nice
high resolution image, although it can sometimes also be useful for improving
the spatial normalization step. If you can get a good match between a set of
rather crumby looking functional images and a nice quality structural image,
then you can estimate the warps for the spatial normalization from the
high quality image and apply them to any of the images that are in register with
it.
We do spatial normalization for two main reasons:
1) Combining data from a number of subjects whos brains are all different shapes
and sizes.
2) For reporting the results in a consistant euclidian co-ordinate system.
> Maybe some of this is answered in SPM99b, but until we get the new MatLab we
> are in SPM96 Land.
The interface for the spatial stuff is very similar in SPM99b and SPM96.
> Finally, you mention "distortion correction" below. Can you tell me more about
> this, is there a package that handles these corrections?
We are currently implementing a set of distortion correction procedures in Matlab
that are based on the unwarping routines of Peter Jezzard et al (written in IDL).
Details of the theory and application of this software may be found in the paper
"Correction of Geometric Distortion in Echo Planar Images from B0 Field Variations"
by P. Jezzard and R.S. Balaban, Magn. Reson. Med. 34, 65-73 (1995).
All the best,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|