Hello,
and thanks for your comments!
It is hard to try explaining something one does not know well oneself,
that is why my questions probably sound not totally sensible.
> This seems odd; in the example that you sent, once I turned off the
> initial registration, the general accuracy I would say was way better than
> 1cm. I'm not sure exactly what steps you are going through to do this
> test?
I did it like this: I have a mask that has been drawn on a subjects
highres image to cover an anatomically defined area. I registered epi
straight to highres using 12 DOF, full search. Then I used the
highres2example.func.mat FEAT produced to transform the mask to the space
of functional images and wiewed the transformed mask as an overlay on
example.func image.
This way the mask ends up to slices 13-16 of the epi-image.
As I have the T1 axial volume taken with the same slice
prescription as epi I know the area of interest is in slices 10-13. Also
judging on the basis of the form of the lateral ventricles one can see
that the transformed mask lies too high.
This is not a great problem any more since I have found one way around it
but not being able to use the initial highres according to my original
plan bugges me.
> Ah, no, I'm saying that the voxel sizes in one or other of the original
> images was slightly _wrong_, not that they should be set to be the same!
> Ie the true voxel size in one or other image was not exactly what was put
> into the header. You will need to use greater than 3DOF to find out what
> the right ratio of voxel sizes should be.
Really - how can that be found out? I think that if there is a discrepancy
with true voxel sixe and the size announced in the header this might need
some scrutinizing. Also, if one can find out the "amount" of discrepancy
could one scale the initial highres with a factor which could correct it?
With best regards,
Johanna Pekkola
LCE/TKK
Espoo, Finland
|