Dear Jess,
> Simple question from an SPM neophyte: We ran six sessions on the same subject;
> each session had a total of 60 scans in a boxcar alternating between baseline
> and a task (a total of 30 scans in the baseline, 30 scans in the task). The
> task in session 1 and 4 can be considered A+B; in 2 and 5, A+C; in sessions 3
> and 6, A+D.
>
> The design matrix is set up with one column per session, each session
> identical, each session modelled as a boxcar, in the order mentioned above.
>
> 1) If I use the contrast [1 1 1 1 1 1] and a cluster shows up, is that finding
> voxels which were significant in *all* sessions, or voxels which were
> significant in *any* session? Is that the best way to see the effects of A
> which are consistent across all tasks?
Neither. This contrast is looking for areas where, *on average* across all
sessions, there is an effect of active task vs baseline. This subsumes
task-by-session interactions. In the worst case scenario, if there was a
gigantic activation in session one, and much smaller activations in all
other sessions, this could show up overall as 'significant'.
If you are interested in identifying areas that are 'significantly'
activated in each and every session, then a conjunction across sessions
would do the trick. This would in effect discount session-by-task
interactions.
> 2) If I want the effects of B without A: should I use [1 0 0 1 0 0] and mask
> it with [ 1 1 1 1 1 1 ], or mask it with [0 1 1 0 1 1]? or something else
> entirely? I thought about [1 -1/2 -1/2 1 -1/2 -1/2], but I didn't think giving
> the non-B sessions negative weights was the right thing to do because I want
> areas which just increase with B, regardless of whether they decrease for C
> and D. Or is there a better approach to pulling out the effects of B? (and
> eventually C and D)
I'm a little confused here. My understanding of the design was that session
1 represents a task which subsumes cognitive components A & B, alternating
with rest; and condition 4 is an exact replication of condition 1. If this
is the case, then none of the contrasts you suggest will isolate cognitive
component B. But I think I may have misunderstood! If instead, condition A
is the baseline condition replicated over sessions, then the first contrast
you suggest would work (without masking).
> 3) Tangential question: If we want to do an analysis only in one hemisphere or
> region of interest, can that be done in SPM99?
Yes - although you would still run the analysis on the whole dataset. The
small volume correction button in SPM99 allows statistical inference on
small volumes of interest. Run the analysis as normal over the whole brain,
and use the SVC button to generate statistical inference for a small ROI
centred on a given location (or specified by another SPM).
best wishes,
Geraint
------------------------------------------------------------------
Dr. Geraint Rees
Wellcome Advanced Fellow, Lecturer,
Mailstop 139-74, Institute of Neurology,
California Institute of Technology, University College London,
Pasadena, 12 Queen Square,
CA 91125 London WC1N 3BG
voice (626) 395-2880
fax (626) 796-8876
web http://www.klab.caltech.edu/~geraint
------------------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|