Dear Martina,
>I use SPM for analysing a MRI working memory task in subjects with AD and I
>am wondering, if someone can help with choosing a cut-off period for this
>task / box-car design with three different conditions. Each condition takes
>30 sec, TR of 3 sec, task-duration 10 minutes.
>The design: Rest (a) zero back condition (b) for 5 minutes; Rest (a) one
>back condition (c) for 5 minutes)
>
>a b a b a b a b a b a c a c a c a c a c
The default settings that SPM will estimate when you are asked for a
high-pass value here are calculated by taking 2x the minimum of the maximum
inter-trial/inter-epoch interval. So, in your case, if you are modelling
each condition separately i.e. one regressor for each condition a,b and c,
the cut-off frequency will be driven by the b and c conditions because,
taken over the entire session, they occur least frequently. Basically,
setting the high-pass at 2x this interval (or merely accepting the default
that SPM gives you) should ensure that your *expected* experimental
variance should not be modelled by the high-pass filter - which is what you
want!
>Or does anyone have an even better idea for analysing this set of data? For
>example to split it into two separate sessions?
I don't think you really want to split this into two separate sessions
unless you have a really good reason to. I would guess that it would have
been nice to randomly intersperse your one-back and zero-back conditions,
but this may not have been possible with your AD patients. However, as long
as half of your subjects begin with the zero-back, and half begin with the
on-back, you should be immune to the worst effects that order and fatigue
may cause.
One further idea that you may consider is that your actual contrasts will
probably have a higher frequency than the frequency of your covariates,
because I'm assuming you will want to examine differences (i.e. b-a and c-a
rather than the main effects of a,b and c per se). Thus you would set your
cut-off appropriately lower (i.e. having a higher frequency).
I should note at this point that there are a number of different views on
how one deterimes the cut-off period, particuarly with respect to whether
one should be most driven by the frequency structure of one's signal (in
this case) or by the frequency structure of the noise. The approach above
is merely the 'vanilla-flavour' approach (or 'vanilla-flavor approach' as
they spell it in my wonderful country of residence).
Best,
Dave.
=========================
David J McGonigle
Biomagnetic Imaging Lab,
Box 0628, Dept. of Radiology
UCSF, 513 Parnassus Ave
San Francisco
CA 94143 USA
=========================
|