Dear Saemann,
I'll have a go on this, I used RT parametric modulation a couple of
times. Hope my comments make sense...
Saemann Philipp schreef:
> Dear SPM users,
>
> we are analysing an event-related fMRI paradigm - four different trial
> types are encoded in the
> first level design matrix, two of those are compared against each other
> using t-tests.
>
> We would now like to attribute reaction times to each single trial,
> assuming that there is a linear
> relationship between BOLD amplitude and faster reaction times (for a
> start). We are still interested in the difference
> between trial type 3 and 4, however, the faster ones should be given a
> stronger weight.
>
I do not understand what you mean with the latter; when using parametric
modulation, you model RT-modulated HRF amplitudes with a second
regressor with HRFs that scale with mean corrected RTs associated with
each individual event. See for example Neggers et al 2005 (21(10),
2853-2863), where I use saccade RT parametric modulation of HRF
amplitude, figure 2 clarifies the appoach. I.e., HRF size variations
following fast and slow reactions are modelled in 1 single regressor (or
2 or 3 when using temporal and perhaps dispersion derivatives).
> Our questions are:
>
> (1) Is parametric modulation the right way to include such behav. measures?
>
When your underlying neural theory assumes increases/decreases in neural
responses associated with faster/slower responses (or reverse), then
yes. It all depends on what you assume the neurons in the system of
investigation do whether this is appropriate...
> (2) If yes, if we assume that faster reactions (lower reactions times) are
> associated with higher BOLD response,
> how is this relationship correctly encoded in the first level DesMtx? (It
> seems that direct use of reaction times flips the relationship...)
>
Doesn't really matter what dependency you are looking for (positive or
negative), the model will be the same. Your contrast of course will be
different;
when you expect higher HRFs with lower RTs (as I did too), you look for
a negative linear dependency between your parametrically modulated
regressor and the variance not modelled by the canonical HRF regressor.
Or, in other words, you then test for the contrast [0 -1] (assuming
there is one event type, 1 modulating factor, no derivatives, adapt
accordingly when using other designs with more factors) to be different
from zero. Use [0 1] when you expect a positive linear relationship.
Things get a little more complicated when modulating higher order
non-linear depencies, but the idea is the same there.
> (3) If we are still interested in T-contrasts (rather than just F-contrast
> based differences), how are these contrasts interrogated? Do we need
> another Design matrix that lacks the binary (unmodulated) regressor?
>
I am not sure what you mean here, perhaps the contrast in my above
comment makes sense...
>
> Thank you very much for any advice here,
>
Anytime
> Philipp
> Max Planck Institute of Psychiatry
> NMR Research Group
> Kraepelinstr. 2-10
> 80804 Munich
> Mail: [log in to unmask]
> Phone: 0049-89-30622-413
>
--
-------------------------------------------------
Dr. S.F.W. Neggers
Division of Brain Research
Rudolf Magnus Institute for Neuroscience
Utrecht University Medical Center
Visiting : Heidelberglaan 100, 3584 CX Utrecht
Room B.01.1.03
Mail : Huispost A.01.126, P.O. Box 85500
3508 GA Utrecht, the Netherlands
Tel : +31 (0)30 2509609 Fax : +31 (0)30 2505443
E-mail : [log in to unmask]
Web : http://www.fmri.nl/people/bas.html
--------------------------------------------------
|