Dear Alex,
>
> I need your help with setting up a design matrix for a sparse fMRI
> study designed to look at responses in auditory cortex.
>
> Subjects heard 10 secs of auditory stimuli with the EPI off. Then the
> EPI was started and collected 3 seconds of info. A 10 second gap
> followed and the next stimulus was given; i.e: Acquisition time (TA)
> was 3secs with an interscan interval (TR) of 20 secs. There were four
> conditions with 10 scans each = 40 scans in all. I'm assuming that this
> is like an event-related fMRI study.
>
> The problems is that I cannot seem to get spm99 to model for the part
> of the BOLD signal I'm looking for. Because the stimuli started 10sec
> before the EPI, there is no initial dip and then rise in my signal
> (i.e. no 'boxcar'), simply the (flattish) peak of the BOLD signal.
The fMRI design routines are configured for evenly spaced data
sequences, whereas your time-series are discontinuous. I think the
best way to proceed (and there are several) is to pretend you have a
continuous time-series with a TR of 3 seconds. Specify a series of 10
scan epochs and model the epoch-specific responses with a small number
of basis functions (e.g. mean and exponential decay or a non-windowed
(Hanning) Fourier set of order 3 to 4). Do not convolve with the HRF
and use F-contrasts to test for mean effecta and differences in the
epoch-related responses. The latter will be the tail end of the
hemodynamic response after the period of stimulation and will probably
show some form of decay (with an undershoot). The critical thing here
is that you should use an interscan interval of 3 seconds and note that
you cannot simply convolve a stimulus function with the HRF. This is
because the stimuli were, from the point of view of the analysis,
compressed into a point in time between the epoch onsets.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|