Christian Matthiae wrote:
> Hi All
> i am doing an dcm analysis and right now i am having some questions and
> hoped someone could help me:
>
> 1. for some subjects i cant do the parameter estimation (at least not in an
> acceptable time), it stops after the first or second E-Step and doesnt go on
> for several days (then i cancled the computation). I observed this for
> several subjects with a model which is estimated for other subjects in under
> an hour. I am using the new spm_nlsi_GN version. Any ideas what the reason
> for this could be?
Convergence usually fails when estimates of parameters become very collinear.
This is usually overcome by simplifying your model. I don't know how many
regions/connections you have but maybe you could reduce them to the
absolute minimum.
If you are a power user (ie. don't mind hacking code) you could do this
in a data-driven way by stopping the
estimation after the first or 2nd step and have a look at the DCM.Cp matrix - the
matrix of posterior covariances. This will tell you which parameters are
highly collinear.
>
> 2. i am not quiet sure about this: In DCM you always do the analysis on an
> single subject level or is there a possibility to do group statistics and
> model some kind of group DCM? Or do u have to do this after estimation for
> single subjects?
>
You can follow the usual two-stage ('summary statistic') approach.
Stage 1: within subject analysis - estimate your connectivity parameters for each
subject. Record contrasts of interest (eg. strength of backward vs. forward
connections) for that subject.
Stage 2: T-test on contrasts from stage 1. Unlike SPM/GLM analyses these won't be
images. Just single parameters.
> 3. In our paradigm the subjects were shown emotional pictures and afterwards
> rated the emotional strength of this pictures. Is there a possibility to use
> this values as a parameter in DCM?
>
Yes - replace the t-test in stage 2 above with a correlation analysis.
Best,
Will.
> Thanks in advance for all answers
> Christian
>
>
--
William D. Penny
Wellcome Department of Imaging Neuroscience
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7475
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|