> I am still trying to implement a longitudinal TBM procedure and I am
> getting confused by some of the steps.
>
> My first question is regarding the processing steps used in the
> following paper:
>
> C M Kipps, A J Duggins, N Mahant, L Gomes, J Ashburner, and E A McCusker
> Progression of structural neuropathology in preclinical Huntington’s
> disease: a tensor based morphometry study
>
> In methods section, it reads: "
>
> 2. We sought a high dimensional deformation field (using the spm2b
> "Deformation Toolbox") that would warp the early T1 image to match the
> late T1 image within subject.34 ...
>
> 3. The result of the high dimensional registration (step 2) is a
> detailed deformation field that contains a mapping from each point in
> the late image, to the corresponding point in the early image, for each
> subject..."
>
>
> Here is my question:
>
> I am confused here .. .if you are warping the early image to the late
> image, shouldn't the result be a deformation field that contains a mapping
> from each point in the early image, to each point in the late image?
What is generated is a mapping from the space of the warped image, to the
space of the original. Generating a warped image involves looping through
the voxels of the image being created, and finding where the values should
come from in the original image. This needs a mapping from the warped to the
unwarped image, or in this case from the late image to the early image.
If it was done the other way around (i.e. looping through the voxels of the
original image and putting those values in the appropriate place in the new
image), then this would leave some gaps (unless it was done in a rather more
complicated way).
Best regards,
-John
|