My best guess is that one of the steps didn't work, and probably got
caught in a local optimum. Figuring out which step failed would require
looking at the data. I suggest you use Check Reg to see if the
coregistration worked adequately, and also to check whether your
spatially normalised images images are in alignment with the templates.
Local optima are problematic for spatial processing, and completely
avoiding them would require a more complete search through all the
possible parameters. So, for a rigid registration, we may try 20
different values for the six parameters, which would give us 64,000,000
different combinations. Each of these combination would need one of the
images to be resliced to see how well it matches.
The SPM community wouldn't be very happy about the time needed to do
this, so we use a different approach. Instead we use a local
optimisation strategy, which relies on the images being roughly aligned
beforehand. Even this can be problematic though, as sometimes the
solution we are after does not correspond with that with the most
optimal objective function. In particular, the coregistration model may
not be good if images are corrupted by bias artefact.
If you want a procedure that is more robust, I you could try:
-----------------------------------------------------
Slice Timing (if you must)
Realign: Estimate and Reslice
NEW Segmentation (of structural scan) to output bias corrected, GM, WM
and CSF.
Imcalc: Use i1.*((i2+i3+i4)>0.5) on the output from the segmentation.
Coregistration: Estimate (source = output from ImCalc, target = mean
image from realignment)
Deformations utility to take the deformation from the new segmentation
and write out the normalised data.
Smooth
-----------------------------------------------------
It might seem convoluted, but the batching system should be able to
handle it easily. Create the job once and then you can use it multiple
times with different input. Simple.
Best regards,
-John
On Wed, 2010-02-10 at 19:01 -0600, Steffie Nelson wrote:
> Dear all,
>
>
>
> I’m running a preprocessing script that involves the following steps
> (modeled after spm8 manual example beginning p.195):
>
>
>
> Slice Timing
>
> Realign: Estimate and Reslice
>
> Coregistration: Estimate (source = structural, target = mean image
> from realignment)
>
> Segmentation (of structural scan)
>
> Normalisation: Write
>
> Smooth
>
>
>
> The problem is this: after processing, I can look at the activity data
> from a particular contrast and the functional activity is clearly NOT
> normalized. I’ve attached a .jpeg of a results graphic to
> demonstrate. Has anyone run into this problem?
>
>
>
> Thanks in advance,
>
> Steffie
>
>
>
>
>
>
--
John Ashburner <[log in to unmask]>
|