Hi Ville,
When I've analysed sparsely sampled data with Feat I've treated it much
like PET data. If the time points are far enough apart, then you can assume
no temporal autocorrelation, making each point independent. In this case,
there is no meaningful time course which needs to be correlated with a
reference wave form. Instead you have dummy variables in the GLM just
marking the various conditions. Obviously it's important that you've
sampled the brain at the correct time period (ie. when the BOLD response is
peaking or near peak in your region of interest -- often auditory
cortex). Just create three different EVs such as:
EV1:
1 1 1
2 1 1
EV2
3 1 1
4 1 1
5 1 1
EV3
6 1 1
7 1 1
8 1 1
9 1 1
10 1 1
Set your TR to 1 (which is arbitrary but it serves as a simple index of the
scan number). You could do the same thing with the 1 column format by
simply dividing each of the columns from your original mail into an EV. As
you say, don't convolve the EV with any HRF and don't include temporal
derivatives.
Note, however, that this solution really works best for a single brain
region because other areas may have different timing of the BOLD response
and will of course be sampled at different times.
Good luck!
Joe
Joseph Devlin, Ph. D.
FMRIB, Dept. of Clinical Neurology
University of Oxford
John Radcliffe Hospital
Headley Way, Headington
Oxford OX3 9DU, U.K.
Phone: +44 (0)1865 222 738
Fax: +44 (0)1865 222 717
Email: [log in to unmask]
|