Dear Charlotte,
> I am planning an fMRI study which I hope will include effective
connectivity analyses, and after reading your 1997 and 2000 papers with Karl
Friston, I still have a few questions, some specific and one general:
>
> 1. Is is necessary that epoch length be a whole multiple of effective TR?
No. That does not matter.
> 2. Are longer epochs (say, 8-10 scans per epoch) a requirement to meet the
assumptions of SEM analyses of condition-specific effective >connectivity?
I notice for condition-specific analyses, you are using essentially an
interrupted time series rather than a continuous one. >However, SEM doesn't
necessarily require time series data at all (and I think treats each time
point as independent).
You do not necessarily need long blocks. The only requirement is that the
hemodynamic responses for different conditions/events do not overlap,
because then you are mixing the sources of covariance and would want some
deconvolution.
SEM does not need time-series, and I agree, temporal autocorrelation needs
to be taken into account (ie effective degrees of freedom).
> 3. More generally, is there any incorporation of the hemodynamic lag into
condition-specific analyses, or do you just use a boxcar function to >select
the scans collected during the condition of interest (or perhaps with some
fixed lag)?
In this case we included a lag of 2 seconds. I would however advise you not
to break time-series in chunks, but rather analyse the whole time-series
and add an additional regressor, that codes the difference between
conditions, or even a parametric effect and assess the interaction effect.
> 4. As an aside, when you include an interaction term to look at modulatory
effects, do you mean-center the time series before multiplying >them? (It
seems that at least within a particular condition, they would not
necessarily be centered; and your graphs seem to suggest not.)
You should center the time series before multiplying to make main effects
uncorrelated wrt the interaction.
-Christian
Dr. Christian Büchel
Neurologische Universitätsklinik, Haus B
Universitäts-Krankenhaus Eppendorf
Martinistr. 52
D-20246 Hamburg
Germany
Tel.: +49-40-42803-4726 Fax.: +49-40-42803-9955
[log in to unmask]
www.uke.uni-hamburg.de/kliniken/neurologie/pages/mitarbeiter/buechel_c.htm
|