Dear Maria,
> I ran SPM99b.
> We have tested 11 normal young subjects with H2 O15 PET
> study in five different
> conditions 1 at rest and 4 with auditory stimulation.
> The first design matrix to check the differences between
> stimulation-no
> stimulation we used is:
> multisubject conditions & covariates
> conditions 0 -1 -2 1 2 (rest, cond 1, cond 2 cond 3 and cond
> 4)
> Ancova by subject
> scale subjects grand means to 50
> analysis threshold 0.8
> mean voxel value (within per image fullmean/8 mask)
> Is the right approach?
> Should be -4 1 1 1 1 ?
Yes, the design matrix specification part sounds fine. You should take
the contrast [-4 1 1 1] to compare the mean effect of all stimulation
conditions vs. rest.
>
> The design matrix we used to look for different activity in
> one of the four
> conditions was.:
> multisubject conditions & covariates
> conditions 0 1 1 1 -3 (rest, cond 1, cond 2 cond 3 and cond
> 4 wich is the
> relevant)
> Ancova by subject
> scale subjects grand means to 50
> analysis threshold 0.8
> mean voxel value (within per image fullmean/8 mask)
> what is the difference using 0 -1 -1 -1 3, assuming we
> espected to find greater
> activity at condition 4?
The difference between [0 1 1 1 -3] and [0 -1 -1 -1 3] is that the first
tests at each voxel, whether the mean effect of conditions 1 to 3 is
larger than the effect of condition 4 and the second tests, whether the
effect of condition 4 is larger than the mean effect of condition 1 to
3.
> Which parameter gives the diferential statistical "weight"
> of one variable, the designed number
> of the condition?
I'm sorry, I didn't understand this question...
>
> I was also interested in measuring rCBF in auditory cortex
> at the different
> stimulation conditions. Unfortunately the differences
> between conditions are so
> subtle that using a T test there is no significance with the
> actual number of subjects (11).
If you want to see the effect of each condition vs. rest, then you could
try the contrasts
[-1 1 0 0 0]
[-1 0 1 0 0]
[-1 0 0 1 0]
[-1 0 0 0 1]
If you have a specific hypothesis that e.g. condition 4 should activate
more than condition 1, you could try [0 -1 0 0 1].
> What is the statistical value of contrast of parameters
> estimates?
A contrast is just a specific weighting of the parameter vector. This is
used to specify a null hypothesis at each voxel (e.g. there is no
activation in condition 1 as compared vs. the rest condition). At each
voxel, a t-value is then computed by dividing the scalar product of the
contrast and the parameter vector by the estimate of the standard error.
After this a p-value (corrected for multiple comparisons) is computed
for each t-score. You can then assess the significance of this p-value
(e.g. p <0.05).
> Is ti possible
> to test that for a cluster not just a voxel?
Yes. The Gaussian random field theory is used in SPM to estimate
corrected p-values (derived from a t-map) for voxel, clusters and sets
(group of clusters). In SPM99, you can see these p-values after pressing
the 'volume' button in the bottom left window. By the way, maybe you
should update your SPM99b version to the latest SPM99 version.
Stefan
--
Stefan Kiebel
Functional Imaging Laboratory
Wellcome Dept. of Cognitive Neurology
12 Queen Square
WC1N 3BG London, UK
Tel.: +44-(0)20-7833-7478
FAX : -7813-1420
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|