Dear Torsten,
> Dear SPMers,
> I use SPM for a short time. So i am not familiar with SPM99 so very much.
> I have got a simple question, i have got one group of 60 FDG-PET scans and
> i try to find out the variance for each voxelvalue within this group?
> Is this possible with SPM99? If yes, how?
>
It may not be what SPM is most often used for, but is is certainly possible
if you trick SPM a bit.
If we assume that you have already spatially normalised your images (using
SPM or otherwise) you can do as follows. In the PET interface select "Basic
Models", in there select "One sample t-test" and then just pick the defaults
(But don't do "grand mean scaling" because that would change the unit of your
variance estimate). SPM will now model your scans using a single mean, which
means that the residual error of the model is simply the variance. One of the
outputs of this will be a file named ResMS.img, and the pixel values in that
image will represent (n-1)*Var, where Var is the variance in that voxel and n
is the no. of scans you put in (i.e. 60 in your case).
The above assumes your images are quantified to yield rCMRglc values. If they
are simply activity images I suggest you do:
PET-Models->Single Subject Covariates Only->0 Covariates->0 niusance
variables->Proportional Scaling etc. That way you will have normalised for
differences in injected activity. You will have lost your original scale
(cause the global normalisation is done to a user specified value) but on the
other hand you had no really defined scale to begin with.
ResMS.img will again be as defined above.
Good luck Jesper
|