I've used FSL for PET analyses in the past and it seems to work fine as
long as you turn off the fMRI-specific features and use a PET-model rather
than the (implicitly assumed) fMRI model.
Your parameter choices all look fine. The only issue really is whether you
want to analyse the subjects individually (as you seem to have specified)
or within a single design matrix. Normal PET practice is the latter and it
rests on the assumption that the intra-subject variance (i.e. between
scans) in PET is a similar magnitude to the inter-subject variance. As far
as I know this is a largely untested assumption but since it is the
standard assumption, there shouldn't be any issue about making it. In
practice it means you have far more degrees of freedom when estimating your
model. In addition, it does include inter-subject variability which you'd
want if you want to make inferences at the population level (i.e. random
effects). The only question is whether it includes the right mixture of
intra- and inter- subject variance, and as I mentioned, this is still untested.
PET analyses typically combine all the data from all the subjects into a
single simple design matrix -- in your case, with 6 conditions. So you'll
need to concatenate the data from all your subjects into a single 4d
file. For these data to be comparable, they'll need to be registered into
standard space first and resampled. In essence, you'll do all your stats
in standard space, rather than doing them in native space and converting
them later.
Since there is no meaningful TR, your choice of 1is fine (in fact, it's
easiest) and use the 3d EV format to specify the scans which correspond to
the individual conditions. As you say, make sure to turn convolution (with
either basis functions or the HRF) off for each EV. Also, don't include
temporal derivatives nor high pass filtering (as PET doesn't have the
issues of aliasing nor temporal drift). Similarly, turn pre-whitening off
because PET also doesn't have the problem of temporal autocorrelation.
This will produce a rank deficient model if you include rest. This
shouldn't be a big issue if all of your contrasts add to zero (ie compare
some conditions to others). You won't be able to look at contrasts like [1
0 0 0 0 0].
Good luck with it,
Joe
--------------------
Joseph T. Devlin, Ph. D.
FMRIB Centre, Dept. of Clinical Neurology
University of Oxford
John Radcliffe Hospital
Headley Way, Headington
Oxford OX3 9DU
Phone: 01865 222 738
Email: [log in to unmask]
|