> today, I have two questions on VBM using SPM5b:
>
> 1.a)
> Is it still possible/advisable to use the following function:
>
> function gl = get_integrals(P)
> % Integrate the values in an image.
> if nargin<1,
> P = spm_get(Inf,'*.img');
> end;
> V = spm_vol(P);
>
> gl = zeros(length(V),1);
> for i=1:length(gl),
> for z=1:V(i).dim(3),
> img = spm_slice_vol(V(i),spm_matrix([0 0
> z]),V(i).dim(1:2),0); gl(i) = gl(i) + sum(img(:));
> fprintf('.');
> end;
> fprintf('\n');
> end;
This function gives the sum over all voxels of an image. These values should
be proportional to the total tissue volume of a tissue probability map image
in native space, or a "modulated" spatially normalised image.
>
> 1.b)
> If so, can I use this function separately for modulated and unmodulated
> normalised GM- and WM-images produced by the SPM5b segmentation step, i.e.
> smwc1*.img, smwc2*.img, swc1*.img, swc2*.img ? Then, I would use the four
> different sets of integrals/globals that will be produced by this function
> as covariates of no interest during the four respective analyses, i.e.
> GM-/WM-vol and GM-/WM-"density".
You could use any meaningful measure as a confound in your statistical
analyses. There is also no reason why you should be limited to using just
one confound either. Basically, it depends on what question you are trying
to ask of the data.
>
> 2.) I realised that, unlike in SPM2, the segmented images (smwc1*.img etc,
> see above) in SPM5b have a resoluation of 2x2x2 mm since the new
> SPM5b-templates and prior probability maps have this resolution. The images
> I start with in native space, though, are 1x1x1 mm. Therefore I wonder
> whether it might be preferable to use templates and prior maps with the
> same, 1x1x1 mm, resolution? Or doesn't it make any difference? I simply do
> not know at what stage during the new integrated segmentation step the
> resolution is decreased to 2x2x2 mm and whether partial volume effects
> could be diminished by keeping it up at 1x1x1 mm.
As the data are usually smoothed afterwards, then it shouldn't make so much
difference - especially as the tissue probability maps that are used as
priors are fairly smooth. In SPM5b, the actual segmentation is done on
images in their native space, so there shouldn't be so many problems
pertaining to partial volume effects. After the images are segmented, then
they are smoothed by a tiny amount (to reduce aliasing effects) and warped
into the standard space.
Best regards,
-John
|