Dear fellow SPM-users,
I had a maybe bit strange idea of which I'm curious whether it is
valid and whether it is doable in SPM.
The idea: I have a design in which I have 18 versions of one regressor
(i.e., same trial type/structure, but different conditions). I want to
estimate my HRF shape using FIR models on a regressor for only trial
vs no trial and then use the estimated HRFs per voxel for the analysis
with the full 18 separate regressors. The latter I could either do on
the same dataset (given that my contrasts are within the 18 regressors
it doesn't appear to be double dipping) or on a separate one (the safe
option). The reason why I don't do it in one go, so just have a FIR
model to start, is that that would give me too many and too correlated
regressors. Is this valid/does this make sense?
The implementation: if it is allowed and actually makes sense, how
could I implement it? I see two ways: either have different design
matrices per voxel, which seems too big a hack, or deconvolve the data
with the custom HRF's and then fit a non-convolved version of the
design matrix. The second is probably the most realistic. I could
deconvolve the data in matlab, write the images back to disk, and then
put all my regressors as regressors (together with the movement
regressors) instead of conditions, to make sure they are not
convolved. Is that the best way to approach the problem, or are there
better alternatives?
I hope this is clear, thanks a lot in advance,
best,
Frank Leone
Graduate Student, Donders Institute for Brain, Cognition and Behavior
|