Karsten -

> In my event-related experiment with auditory stimuli (triggered by the
> scanner),
> i have found, that there are a few significant voxel correlates with the
> assumed hrf, but
> much more in the auditory cortex with the temporal derivatives, in all
> subjects.
> 
> So, there are two questions.
> 
> 1. Is the temporal delay of the BOLD response included, when calculating
> the hrf, or
>    must i add the delay manually?

The temporal delay of the BOLD response is included in the canonical
HR function assumed by SPM.


> 2. If the delay is calculated by SPM, so what does it mean, when only
> the temporal
>    derivatives shows high significant results. Is the auditory system
> faster then the
>    synthetic hrf, because in an experiment of the visual cortex, the hrf
> fitted the data
>    very good.

This is a good question, to which there are several possible answers.
One possibility is that the stimulus times you entered for your
analysis were slightly out. As mentioned in a previous email by
Christian Buechel, if your stimulus times are variable, the times you
enter into SPM are in units of scans, starting from 1 for time t=0 at
the onset of the first scan. Thus if, for example, your stimulus
times, in seconds, were stored in a vector V, the stimulus times
required for SPM would be (V/TR)+1, where the constant TR is your scan
repetition time (in seconds). These would be the correct stimulus
timings for the first slice acquired.

As mentioned at the end of my previous email, another possibility is
that the above stimulus times are inappropriate for the slice
containing the part of auditory cortex in which you are
interested. This can arise if you have a long TR and the slice of
interest is acquired relatively late in the sequence (eg >1 seconds
later than the first slice; assuming that you have not performed some
form of interpolation in time to correct for such slice-timing
differences). The Taylor expansion described in my previous email
holds less well for values of dt > 1 second. One possible solution is
to adjust your stimulus times in order to synchronise them with the
slice of interest. So if, for example, the slice of interest were the
j'th of N slices acquired, you would subtract the constant j/N from
the stimulus time vector above. The canonical HRF would then have the
correct delay for your slice of interest, and you might notice that
the canonical HRF now fits better than its derivative. Of course the
timing is always slightly out for the remaining slices, but remember that
one purpose of the derivative is to cater for small time differences
of about a second. Thus assuming that you took the middle
slice acquired as the reference slice (by subtracting TR/2 from your
stimulus times), this problem only really arises for TRs > 2 seconds.

A final possibility is that your timings are correct and the brain's
event-related response really does occur earlier in time than
predicted by the canonical response assumed by SPM. On the basis of previous
studies performed here however, in which responses in auditory cortex
have been well-fitted by the canonical form, this would not be my
first guess. One way to test this would be to use a more general basis
set, such as a fourier set, which is more robust to slice-timing
differences, plot the best-fitting function, and see where its peak
occurs in time.

Sorry for the long-windedness; I hope this helps,
Rik.



-------------------------------- 8-{)} ----------------------------------

DR R HENSON                         INTERNET    [log in to unmask]
Wellcome Department of              
 Cognitive Neurology                TEL (work1)    +44 171 833 7483
12 Queen Square                     TEL (work2)    +44 171 391 1138
London, WC1N 3BG                    FAX            +44 171 813 1420

        URL: http://www.psychol.ucl.ac.uk/rik.henson/index.html

-------------------------------------------------------------------------



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%