Hi Jack,
It's hard to say why it would behave like this, especially as it only
goes
beyond 7 dof in the last bit of the 2mm and 1mm resolution optimisation.
This behaviour normally indicates that there is something significantly
mismatched between the images (say bias field or distortion artefact)
which will try to drive one of the individual scalings away from a good
value and if it goes far enough then everything else goes haywire.
If the images both look good then I dont know why it would happen.
If you like, you can make them available for download and I'll have a
look.
As for the last part - about example_func to highres - bias field
correction
on an EPI image is normally not very important (as the voxels are
normally
pretty big) but if you do see something obvious then yes, try doing it
with
FAST. Bias field correction on the structural images is normally more
important and works better (as FAST can get better segmentations and
bias field estimates). If the bias field is obvious then it is better
to remove
it if possible.
All the best,
Mark
On Thursday, March 4, 2004, at 01:24 pm, Jack Grinband wrote:
> Hi Mark,
> I tried registering two hi-res, high contrast images (SPGR
> 256x256x124, skull-stripped) from
> the same individual acquired a few months apart. When I used 7 dof,
> the registration was
> perfect. However, when I used 12 dof, the registration was off by
> about 90 degrees and badly
> scaled (around a factor of 1.5). The results were the same for
> correlation ratio and mutual
> info. I checked the headers, and the voxel sizes were identical and
> correct. Any idea why this
> would happen?
>
> Do you think that applying a bias field correction to example_func
> would improve the
> registration to the hi-res structural image?
> thanks,
>
> jack
|