dear monica,
i think will penny has answered a similar question previously. here is a
copy of that correspondance (see below your message). i hope it helps,
zoltan
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of MONICA GIMENEZ NAVARRO
> Sent: October 24, 2005 5:53 PM
> To: [log in to unmask]
> Subject: [SPM] "one subject with a sample" comparison
>
>
> Dear SPMers,
>
> I'd like to know which is the procedure/steps to do a "one
> subject with a sample" comparison analysis, to compare
> individually one subject to a global sample, by VBM analysis.
>
> Thank you in advance,
>
> Monica
----------------------------------------------------------------------------
> Dear SPMers,
>
> First of all I am sorry if my question is too routine for you, but I
> am really need your answer. We are working on memory and we did an
> event related fMRI study in which were involved 10 subjects. The fMRI
> study consisted on two tasks: memory vs decision (decision was the
> control task). In this moment our aim is to compare this group with
> one patient with memory deficits, and to determine what areas are
> significant activated by the group vs patient and by the patient vs
> group. First of all we took all the 10 cons (memory vs decision) from
> our group and put it in a two sample t test with one con from the
> patient on the other side. So, in our two sample t-test we had 10 cons
> on one side and one con on the other. When we asset the results there
> was no activations for group vs patient, neither patient vs group.
> Is it correct to perform a two sample t test with 10 cons on one side and
> just one on the other?
Yes, that's fine. You can test to see which voxels show significantly more
activity in the patient than in the group using the [-1 1] contrast (or [1
-1] depending on your labelling).
When you say there were no voxels showing any difference what style of
inference were you using: (i) FWE, p=0.05 (ii) p=0.05 uncorrected or (iii)
p=0.001 uncorrected ?
I ask because (i) can be quite conservative.
It may be wise to use a less conservative method eg. (ii) or (iii) just to
see whether any blobs you get are in the right places. Alternatively you
could look at the con images to see if the effects have the right sign - ie.
in places where you expect a decrease you get a decrease.
Also, a lack of a difference in activation, of course, could be due to
(a) a lack of the expected activation in the group or (b) unexpected
activation in the patient.
To find out the cause you should check that the expected activations in the
group are significant at the random effects level (ie. one sample t-test,
that is you are testing against zero rather than the patient activation) -
ie. look at (a).
Also check (b) at the fixed effects level.
Other problems to look out for are that the patient data has been spatially
normalised properly, especially if the memory deficit is due to lesion.
Hope this gives you some ideas.
Best,
Will.
|