Dear Boris,
This issue is not restricted to DCM analyses of connectivity. The Sonty
07 paper is one of my favourite examples of the situation in which
connectivity differences are seen without activation differences (in
terms of classic SPM analysis of regional activation). But a similar
effect has been seen with SEM (Rowe, Brain 2002) or simple correlations
(Rowe, J Neurosci 2007).
In August 2007 there was a very helpful exchange on the SPM list between
Klaas Stephan and Barry Horwitz on this topic (31/08/2007 - 05/09/2007
subject: "DCM - VOIs and clinical groups").
Best wishes,
James
> Dear Boris:
>
> It is 'legal" and may even be ideal for DCM analysis in the sense that
> group differences are not explained by regional responses but rather
> by the influences between regions (Sonty et al, J Neurosci, 2007).
>
> One can consider the standard fMRI convolution model as a special case
> of DCM in which the effects of coupling between regions is discounted
> (Friston, Neuroimage, 2003). In a sense the standard method of SPM
> analysis would be like only including direct effects on each region
> and observing the responses. The result you are getting suggests that
> regions are responding in similar ways between the two groups, but
> that the reason for those responses may differ because of connectivity
> influences, e.g., each group may use different connections or weight
> the connections differentially to achieve the same activations. (I am
> using the term connections in the sense of influences between regions
> rather than strict anatomic connectivity.)
>
> -----
> Darren Gitelman
>
>
>
> On Fri, Oct 17, 2008 at 4:57 AM, boris suchan
> <[log in to unmask]> wrote:
>
>> Hi all,
>>
>> I have a short question about the condition for a dcm analysis.
>>
>> I have 2 groups patients and normals which do not show any activation
>> differences in a "normal analysis".
>>
>> Using dcm, I find some differences which makes sense and verify my
>> hypothesis. Is this legal even if the results from the two sample activation
>> t-test do not show any differences?
>>
>> Many thanks
>>
>>
>>
>> boris
>>
>>
>>
|