The experts may have better ideas, but I have two tentative solutions
that seems to be working so far.
The first way (which I find more statistically appealing) is to
create a mask of the A-only contrast at the group level. To do this,
perform the one-sample t-test on the A-only con* images. After you
have performed the contrast at the group level with the desired
p-value and voxel thresholds, write-filter the image. Convert this
image to a binary mask in ImCalc by loading the image then making a
new image with i1>0 (this may not be required, but it does not hurt).
Then set up a new one-sample t-test for A>B, and when it asks for an
explicit mask at the model stage, use the binary mask from the A-only
test. This process limits your search volume to only those voxels
that were in A-only. I have used this to reduce activations due to
negative activation in B.
The second way is to mask the A>B contrast with the A-only contrast
and write-filter a new image. This is cumbersome, and I have some
questions about how to make the new write-filtered image consistent
with the con* images typically used in the random-effects model, but
it should produce similar results to those above. I still have to
test it further.
Hope this helps.
Amy
At 5:16 PM +0900 1/10/01, Kayako Matsuo wrote:
>Dear SPMers,
>
>I would like to ask how to call out images for masking in the contrast
>manager at the second-level analysis (a random-effect model) in SPM99.
>
>I conducted an experiment that contained two task-conditions, task A and
>task B, in one fMRI session. I made "con*.img" files of 4 contrasts,
>namely, A (1 0), B (0 1), A > B (1 -1), and A < B (-1 1), for each
>participants (n = 12). Then, I performed a one sample t-test of "Basic
>models" for each contrast. I found inexplicable activation in A > B
>statistics, that is, activation foci that were not observed during the
>A-only statistics. Thus, I suppose I have to mask the inexplicable foci
>appropriately.
>
>My idea is that I have to mask the results of "A > B" by the results of
>"A", using the "exclusive" option of "nature of mask". However, I could
>not call out the images other than the contrast for "A > B" and an "all
>effects" at the contrast manager of "A > B" statistics.
>
>Would you show me how to call out the A-only contrast image in the
>contrast manager?
>Or, is my idea wrong?
>
>Sincerely,
>
>Kayako Matsuo
>
>
>*.....
>Kayako MATSUO [log in to unmask]
>
> Electrotechnical Laboratories, Medical Vision Lab.
> OSL Unit, AIST
> 1-8-31 Midorigaoka, Ikeda, Osaka, 563-8577 Japan
> tel) +81-727-51-8424 ext.9831
> fax) +81-727-51-8387
> .....*
|