Hi Ged,
Thanks for your reply and sorry for the confusion. I will describe what
exactly I want to do:
I want to use all preprocessing steps from SPM up to normalization.
Instead of normalization of SPM, I would like to use another
registration algorithm and use the results for SPM-based fMRI analysis.
My registration algorithm requires the input image (source) and the
template to have 256x256 size in XY plane with the same resolution. I am
a bit confused at the coregistration step:
My structural data is 224x256x176 (1x1x1mm). My template is 256x256x185
(1x1x1) and the fMRI EPI data is 64x64x26 (3.3x3.3x5.0).
At the coregistration step, I will register structural to mean fMRI but
I am not sure what option I should select at this step: coreg only, or
reslice and coreg? First, I want to put my EPI and structural in the
same space (coreg) and then use my non-rigid algorithm to register the
structural to the template and use the output deformation fields to
bring my fMRI data to template space as well. So the fMRI and structural
should be both resampled to have 256x256 dim in XY plane and the same
resolution as the template (1x1x1).
When I choose coreg only (at coregistration step) and then resample the
data using reorient (John's code), the output will be 178x233163. I
assumed if we do NOT choose reslice, all the transformation info will be
saved in the .mat file. However, I realized after resampling the data
the center and the size of the structural is not the same as fMRI mean.
Since the centers and the sizes are not the same, I believe I won't be
able to apply the same deformation field (from registration of
structural to template) to EPI images.
Second time, I choose coreg and reslice. This time after resampling the
data (using reorient), the size and the center of the structural is the
same as EPI mean. So I use Ged's code for resizing and changing the
resolution and then apply my registration and use the same procedure for
applying the deformation field to EPI. The only problem is that I lose
info by reducing the size and resolution of the structural. So one way
would be to register fMRI mean to strucutural. There are two problems:
First I am not sure if it would be correct to resample fMRI data to
larger size and finer resolution. Second, I have to reslice individual
EPI images after coreging them with the structural which will take ages.
I really appreciate any help, Thanks.
Regards,
Amir
Ged Ridgway wrote:
> Hi Amir,
>
>> I noticed that when I pad the image using this code or if I change
>> the resolution, the image origin will be changed. In changing
>> resolution case, the actual image size in mm changes as well.
>
> Perhaps I'm confused, but I think these things must happen, in order
> for the image to remain correctly positioned in world-space.
>
>> For instance, if the input image size is 179x233x163 and the origin
>> is at 92.0,95.0,61.0 (spacing 1x1x1), by applying the padding of
>> [-77/2 -23/2 0;77/2 23/2 0]+world_bb(), the size is changed to
>> 256x256x164 and the center is changed to 131.0,107.0,61.0.
>
> This "origin" is the voxel location of the world-space origin (0,0,0),
> since the number of voxels change (this is the padding you want,
> right?) the voxel indices for this location must change in order for
> the world origin to be in the same (world/anatomical) position.
>
> If you view the original Image, and the resized rImage, in SPM's check
> reg, you should find they are anatomically registered -- I don't think
> this can be achieved with what you want here:
>
>> Is there anyway to keep the origin and size the same as its original
>> version? Thanks a lot.
>
> Perhaps you could clarify exactly the behaviour you are looking for?
>
> Best,
> Ged.
>
>
--
________________________________
Amir M. Tahmasebi
Ph.D. Student,
Medical Image Analysis Lab.
School of Computing,
Queen's University,
Kingston, Ontario
CANADA
K7L 3N6
________________________________
Tel: (613) 533-2797
Web: www.cs.queensu.ca/~tahmaseb
|