Dear Claudia,
Let's start from the assumption that the fMRI
response is a convolution of a neural response with a hemodynamic impulse
response function. I'll also assume that you are modeling each event's
response with a single basis function which does not integrate to
zero (that is, you could be using additional basis functions to model that
event, but they would integrate to zero). Third, I still use SPM99, so the
variable names I mention might no longer be appropriate for SPM 2 or 5, but the
theory would still hold.
First, you need to determine the integral of this
basis function for each event type. This is better done using the basis function
representations stored in "Sess" (which can be obtained by loading the SPM.mat
file of SPM99 into the MATLAB workspace) than by using the predictors from
the low-temporal resolution design matrix. In SPM99, the integral (to within a
scaling factor which is the same for every basis function in the
design) for the non-zero integrating basis function corresponding
to event i from session j can be obtained with the following
code:
Sess_j=Sess{j};
integ_event_i_Sessj=sum(Sess_j.bf{i})
Then it must be decided whether it is of
interest to compare between events the integrated neural response OR
the neural response averaged over the duration of the events. To help in making
this decision, this question is equivalent to asking whether the area under
the curve of the neural response or the amplitude of the neural response,
respectively, is of interest. I'll tell you that the neural response amplitude
is the analogue (making lots of assumptions, though) of spikes/sec in single
cell electrophysiological data.
To compare integrated neural
responses , the contrast weights for events a and b within session j would
be
[integ_event_a_Sessj
-integ_event_b_Sessj].
To compare neural response amplitudes, the contrast
weights for events a and b within session j would be
[integ_event_a_Sessj/duration_event_a
-integ_event_b_Sessj//duration_event_b].
I hope this helps,
Eric
----- Original Message -----
Sent: Wednesday, June 01, 2005 5:44 AM
Subject: [SPM] variable duration + scaling of
regressors
> Dear SPM community,
>
> I know these issues have been
adressed before, but I am still not sure how to deal with
> a particular
problem.
> I want to compare trials with variable durations due to
different delay periods (2.5s vs.
> 6.5 s).
> After convolution
with the hrf the regressors for these two trials habe different
>
amplitudes which results in much higher beta-values for the short compared to
the long
> trial. If I want to compare both conditions at the second
level, the chance that I get
> significant activation for the comparison
short-long is much higher the for the
> comparison long-short.
> I
think what I want instead is to compare the integrated amount of activity
between both
> trials. Is there any way to do this?
>
> Or
is it possible:
> 1) to scale the regressors to have an equal
amplitude?
>
> 2) to scale the con-images after the first level
analysis?
>
> 3) any other idea or theoretical
consideration?
>
> 4) Please let me know if I am completely off
track
>
> Thank you for your help,
>
> Claudia
>
>
>
>
> ______________________________
>
> Claudia Preuschhof
> Berlin NeuroImaging Center
> Charité
Campus Mitte
> Department of Neurology
> Schumannstr. 20/21
>
10117 Berlin, Germany
> Phone: ++49-30-450 560 195
> Fax:
++49-30-450 560 952
> ______________________________