Dear Pierre,
>>I am trying to understand the differences between modelling connectivity
>>by the use of SEM or DCM. To date, I have only used SEM and I dread to
>>make a wrong use of DCM.
>
>Generally, you may find the following paper helpful in which we directly
>compared SEM to DCM:
>
>Penny WD, Stephan KE, Mechelli A, Friston KJ
>Modelling functional integration: a comparison of structural equation and
>dynamic causal models.
>Neuroimage. 2004;23 Suppl 1:S264-74.
>http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=pubmed&cmd=Retrieve&dopt=AbstractPlus&list_uids=15501096&query_hl=1&itool=pubmed_docsum
I have just a few comments to add to Klaas'. DCM has many advantages over
SEM, many of which are clearly brought out in the Paper that Klaas cites
above. However, SEM still has advantages in some situations. For example:
1. multiple subjects can be included in a single SEM, enabling one to look
explicitly at between-subjects differences in connectivity (e.g. Mechelli
et al Neuroimage. 2002 Nov;17(3):1459-69. ).
2. SEM allows one to use one region as the moderator of connectivity
between two others. This is not possible in the current version of DCM
(but is likely to be possible in future versions?).
3. the computational demands are very different: a complex model with 20
subjects might take 20 seconds to solve for SEM, but takes nearer 30
hours for DCM on my system.
4. anectodally, the results of SEM for a given connection in a model seem
relatively robust against small changes in remote parts of the
network. DCM seems to be much more sensitive to changes in the model, even
at remote parts of the network. This stresses the importance of model
selection, both anatomically and in terms of the moderator
variables/bilinears. Fortunately, a recent paper by Lee, Friston and
Horwitz (<http://www.ncbi.nlm.nih.govjavascript:AL_get(this, 'jour',
'Neuroimage.');>Neuroimage<http://www.ncbi.nlm.nih.govjavascript:AL_get(this,
'jour', 'Neuroimage.');>. 2006 May 1;30(4):1243-54) is reassuring on this
matter.
Although I expect that DCM will largely replace SEM for connectivity
modelling because of its many advantages, there will be a role for SEM for
some time...
Best wishes,
James
|