Dear Tali and Anna,
>Probably we did not explain ourselves properly. Lets say we have two
>sessions.
>In each one of them there is one condition (A in the first session and B
>in the second),
>and the baseline conditions in each session. We would like to find the
>voxels that are
>activated in A, but sre not activated in B. Namely, voxels that are
>affected by the task
>in A, but are not affected by the task in B, so in B they behave like in
>baseline condition.
>Is it possible to extract such voxels in SPM99?
I think that this was more-or-less how I understood your question
last time, so you may be able to find what you need in the answer I
gave then. The reason why I was raising questions about the design
is that normally if you are interested in comparing the response to A
with the response to B then one would randomize these (or at least
interleave them) within the same session, to avoid confounding the
effects of interest with session effects. Having them in separate
sessions introduces problems of interpretation as I am sure you are
aware, but presumably there is some reason why you were forced to do
this.
How you set about answering your question depends on exactly what you
mean by it. Normally one would try to identify voxels which are MORE
activated in A vs its baseline than in B vs its baseline. Here you
would simply model condition A with one box-car and condition B with
another box-car (convolved with the hrf usually), estimate the model
and compare the two betas using a +1 -1 type contrast. But just
doing this doesn't actually tell you anything about the absolute
values of the betas, just about the relationship between them.
Strictly speaking, to approach the answer to your question you would
need to set a threshold of significance, above which you consider
there to be 'activation', and then test for activation by A (with a 1
0 sort of contrast) and for activation in B (with a 0 1 sort of
contrast), and then negatively mask to identify voxels which show up
in the first comparison but don't show up in the second.
Even this, though, doesn't really answer your question (because this
definition of what constitutes 'activation' is not very biological).
You can't directly test the hypothesis that a given voxel is 'not
affected' in condition B. If it doesn't show up in your 0 1
contrast, this may mean that it is 'not affected' or it may mean that
it is affected but that the effect doesn't exceed the statistical
threshold which you have set (because it is small, or because the
noise is great, or because your model isn't very good.... or
whatever). Classical statistics (as used in SPM99) is not really
designed to demonstrate the absence of an effect.
I hope this, or my previous e mail, tells you what you need!
Best wishes,
Richard.
--
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology,
Institute of Neurology, Darwin Building, University College London,
Gower Street, London WC1E 6BT.
Tel: 0207 679 2187; e mail: [log in to unmask]
|