I'm guessing that you estimated the transform by registering the fMRI
with the EPI template, and then applied the transform to both fMRI and
anatomical data (top left, and centre left). For this to work, the
anatomical should be in registration with the fMRI (ie do all the
within-subject processing first, before you do the inter-subject
registration). Try Check-Reg with an original fMRI volume and the
anatomical scan. If these are not in register with each other, then the
spatially normalised versions won't be either.
Best regards,
-John
On Thu, 2009-12-10 at 10:19 +0100, Andre Szameitat wrote:
> Dear SPMers,
> for a couple of my datasets the normalisation procedure results in bad
> fits (see attached image for an example). Note that our original
> functionl data looks like pretty standard whole brain scans (23 axial
> slices, 192mm FOV). In the attached file, the top-left image is the
> first normalised functional scan, second row to the right is the SPM EPI
> template, last row is the T1 template. That's what I did:
> Slicetime correction -> Realign & Unwarp -> Roughly match MNI space
> (set origin to AC, translate/rotate brain via Display->Reorient) ->
> normalise (Est & Write).
>
> The images we put into the normalisation still look fine, however, the
> output looks like on the image. For one or two subjects it worked to use
> another EPI image as input to find the normalisation parameters.
>
> For the remaining two subjects with bad normalisation results, I tried
> to slightly reorient the images again (translating by a couple of mm,
> chaning pitch by ~.05). My idea was to provide the normalisation
> procedure with a different starting position. However, this didn't help
> (output looked different, but still messed up).
>
> Probably this issue has been raised before, but I totally have no time
> to search the archives and I need a solution until next Monday, since
> the next course on which I analyse these data with students is next
> Tuesday morning...
>
> Any help is greatly appreciated.
>
> Best Regards,
> Andre
--
John Ashburner <[log in to unmask]>
|