> We have been using the modulation step in VBM for the last step before
> segmentation. With spm99 we would modulate the data after segmenting it in
> the last step of VBM. We figured that with spm2b we could save a step and
> modulate the data during normalization before segmentation, but that
> doesn't work for some of my images and now we realize is actually different
> than what we did before. Should we be normalizing the images after
> segmentation like in spm99 using spm_preserve_volume? Does that script
> work in spm2b?
The segmented images should be modulated, rather than the images
that get segmented. This is what the
spm_write_sn(VT(1),prm,'modulate');
bit does. It is the SPM2b equivalent of spm_preserve_volume.
If the spm_write_sn(V(i),prm,dnrm.write) part includes modulation, then
this may cause problems for the segmentation.
If the pre-processing involved segmenting before spatially normalising,
then this would save a couple of steps. Unfortunately, I suspect that
this is suboptimal because the segmentation routines work best on
spatially normalised data, as it allows the prior probability images
to be overlaid better. If un-normalised images are segmented, then only
an affine registration is done in order to overlay the priors. This
is not as accurate as overlaying the priors on non-linearly warped images.
> And how can I look to see if Inf and -Inf values are in my image?
If you are using the script containing:
VN = spm_write_sn(V(i),prm,dnrm.write);
Then insert a line that says:
disp([min(VN.dat(:)) max(VN.dat(:))])
Best regards,
-John
> The only way I have so far been able to replicate creating an image with
> such problems is by including Inf and -Inf in the image that is written.
> The spm_write_vol.m function does not check for these, so it messes
> up the scalefactors etc. I can't figure out why such Inf and -Inf
> values
> should arise though.
>
> Are you sure that modulating the original normalised images is the right
> thing to do? This is likely to cause problems during the segmentation
> stage. If the images were initially fairly uniformly scaled, then the
> modulation will introduce a lot of smooth intensity non-uniformity,
> making
> segmentation slightly harder.
>
>
> The "preserve total" or "preserve concentration" options relate to the
> modulation done during VBM. If spatial normalisation doubles the volume
> of a region, then the "preserve total" option will reduce its intensity
> by a half so that the total amount of signal is preserved.
>
> Consider a multi-subject fMRI experiment where all subjects uniformly
> activate a region of 10 voxels such that an activation causes a signal
> increase of 1 unit. After spatial normalisation, these 10 voxels will
> be shrunk for some subjects, and enlarged for others. If the signal
> concentration is preserved, then after smoothing, the subjects that
> have the shrunken region will have lower activation signal than the
> subjects with an expanded region. This is probably not what people want
> from their analyses.
>
> The preserve total option is an alternative that should allow signal
> to be preserved. I don't know if this option has a long term future,
> but I can envisage situations in which it maybe useful.
--
Dr John Ashburner.
Functional Imaging Lab., 12 Queen Square, London WC1N 3BG, UK.
tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381
fax: +44 (0)20 78131420 http://www.fil.ion.ucl.ac.uk/~john
|