Hi Philipp
>
> Hello FSL developers and users,
>
> - We have performed a pharmacological challenge experiment with
> (randomised)
> verum/placebo design and anticipate network changes in the verum group
> (about) with about the same course. This led us to the idea to use the
> Tensor PICA tool in MELODICA. In fact, in an analysis of the verum
> group
> drastic peaks appear round the expected time.
>
> - How could the placebo group now be integrated into this analysis? Is
> t-PICA suitable (as placebo experiments will rater not show effects
> with
> that time course)?
>
You simple add these as additional data sets in your setup. As far as
I can tell there are 3 possible ways in which differences show up:
(i) There are strong differences wrt the spatial characteristics :
multiple components should then show the difference. Dependent on the
amount of overlap this might be a clear separation into component A
swhowing an effect fro group I and component B,C,D showing an effect
for group II _or_ it might be that one component picks up similarity
whereas another set of components ecpress the differences. What
exactly is going on whould be clear from looking at the time courses
and the subject modes
(ii) The difference is largely in the temporal characteristics
The melodic derived time course is the first EIgenvariate of all the
associated time courses - by looking at the full set of time courses
(in tXX.txt within the /report dir) it should become clear what's
going on
(iii) both groups show effects in pretty much the same network with
very similar temporal dynamics - though one of the group will exhibit
a drop in average strength compared to the other:
This should show up in the 3rd (subject) mode as a significant
difference in mean strength between the groups. You can test for this
in melodic by specifying design and contrast matrices in the GUI
> - A more specific question is how the displayed time course relates
> to the
> individual time course of the above-mentioned components? Is it a
> kind of
> best fit?
>
It is the first Eigenvariate (_not_ the mean!), i.e. the time course
that best explains the largest amount of variance - see the technical
report TR04CB1 (available athttp://www.fmrib.ox.ac.uk/analysis/
techrep/) for details
hth
Christian
|