Dear Dr Kuo,

>In SPM 99, there is an option 'plot' for time activity ploting.
>Afetr choosing 'Event/epoch-relared response' icon, more choices for ploting
>type were listed.
>There is a kind of ploting called "fitted response and PSTH".
>What does "PSTH" mean and how to interpret the difference between the curves
>plotted?

PSTH stands for post-stimulus time histogram.  It gives you the
average BOLD signal within a given time window after each occurrence
of the event throughout the whole experiment.  The interpretation is
not entirely straightforward, but basically it gives you an idea of
the waveform of the response, and thus whether your model for it is
reasonable.
>
>In our study, there are two event types of stimulus category randomized in a
>event-related fMRI protocol.

In this example, provided after each occurrence of event A there is
an equal probability of A and B occurring, and the same is true of
event B (i.e. there is a 'uniform transition matrix' - I think that
this is the right piece of jargon), then the difference between the
PSTH curves for A and B really should reflect differences in the
responses to these two types of events.  This is the approach to
event-related fMRI pioneered by Dale and Buckner, I believe, and it
has the advantage that you don't have to make assumptions about the
time course form of the response (except that you assume that it
occurs within a certain window of time).  One disadvantage is that
you have to have a uniform transition matrix in your design, which is
quite a tight constraint in more complex designs.

>After image preprocessing, the images were
>input to the statistic analysis procedure of event-type analysis. HRF with
>time derivative was selected for modelling. Interaction was not specified in
>the statistic analysis. Then, one simple question: how to make contrasts? 4
>effects I am really interested in, including effect of category A,  effect
>of category B; differential effects of A-B and B-A.  Should I specify the
>contrasts in t-test or F-test ? And what is the difference?

If you want to identify voxels where the amplitude of the response to
event A is significantly greater than zero, then a t contrast (1 0 0
0) would tell you this.  This tests whether the beta for the hrf
regressor for event A is significantly greater than zero.  If you
simply wanted to know if the beta was significantly different from
zero (in either direction) then you could use an F contrast (1 0 0 0)
to tell you this; it's the same as a two-tailed t test in this
situation.  Similarly for event B, obviously.

If you want to identify voxels where the pair of regressors which you
have constructed for event A (the hrf and its 1st temporal
derivative) contribute significantly to explaining the actual data,
then the following F contrast will tell you this:
1 0 0 0
0 1 0 0

This should not be confused with the F contrast 1 1 0 0, which would
tell you if the average beta for the first two regressors is
significantly different from zero, which is unlikely to be of
interest to you.

>Because the
>experimental is in event-related fMRI design type, can I specify a contrast
>for the differential effect between A and B by A-B(for example [1 0 -1 0])
>in t-test? And then, the first part of question come.

This is exactly what you can do.  Note that you have assumed that the
hrf is modelling the amplitude of the response, and the 1st temporal
derivative is modelling the delay of the response (which in this case
you are treating as a confound).  This is fair enough, and I think
that most people would consider this to be a reasonable assumption.
However, it is not inconceivable that, in some areas of the brain,
the time course of the real response may not conform to the
'standard' hrf (and might, if you were unlucky, even look like the
1st temporal derivative!) in which case the simple interpretation
that regressor 1 models the amplitude and regressor 2 models the
delay may not be exactly correct.

>Thnaks for your time in advance.
>Nissen Kuo.

No problem.  I hope you don't mind my copying this reply to the
discussion list.  That way other people can check that what I have
written makes sense!

Good luck with the analysis,

Richard.
--
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology,
Institute of Neurology, Darwin Building, University College London,
Gower Street, London WC1E 6BT.
Tel: 0207 679 2187;  e mail: [log in to unmask]