Hi all,
I'm trying to understand the relation of the FEAT design.mat file, the PE
images and the residual time series. I'd be glad if someone could comment
on the following thoughts:
design.mat contains the models of all real EVs convolved with the
respective HRF function and sampled for each timestep in my functional
dataset.
So if I have a design with two original EVs and use the Gamma HRF with
added temporal derivatives, I'll end up with four column vectors in
design.mat
These models are fitted (all together) to my (filtered) functional timeseries.
What I get are parameter images for each real EV (four in my example) for the
best fit of the model to the data..
The residual timeseries (res4d) is what cannot be explained by the full
model fit ( filtered_timeseries - full_model_fit ).
If the above would be true I should be able to recalculate the original
timeseries using the sampled model (from design.mat), PEs and the
residual timeseries.
For a given timestep t and a voxel v this should be:
orig(t,v) = SUM<for all EVs i>( pe(i,v) * m(i,t) ) + res4d(v,t)
where m(i,t) is the value of the sample model for EV i at timestep t.
I tried to confirm this by calculating the equation for some datasets,
but failed to do so. Obviously, there must be something wrong -- either
conceptual or numerical.
I'd be glad if someone could point me to the problem.
Thanks in advance,
Michael
--
GPG key: 1024D/3144BE0F Michael Hanke
http://apsy.gse.uni-magdeburg.de/hanke
ICQ: 48230050
|