Dear SPM experts,
we are encountering an unexpected behavior with smoothness estimation in doing
a RFX on a fMRI data set. The time series of 12 subjects were all smoothed
with an 8 x 8 x 8 gaussian kernel. Different contrast estimates were assessed
at the single subject level yielding a set of con_00* images.
The con images were entered in a set of one-sample t-tests at the second
level, one for each contrast.
Oddly, while for some contrasts the smoothness estimate of the one-sample
t-test lies close to the smoothing kernel (eg. FWHM = 8.3 8.6 8.2 mm), for a
number of other contrasts the smoothness estimate is very low (e.g. FWHM =
1.6 1.7 1.6 mm). For the latter group of contrasts, the activation maps look
quite grainy.
The contrasts with low smoothness estimates seem to involve a subset of
experimental conditions / regressors (i.e. only the contrasts involving
conditions C and D, and not the contrast between conditions A an B).
We have checked the smoothness estimates for each subject at the first level
and they are all in the same range (i.e. approx. 8 x 8 x 8 mm)
Does anyone have an explanation for this behavior?
Thank you in advance for any help!
Marco
--
Marco Tettamanti, Ph.D.
Department of Neuroscience
Scientific Institute San Raffaele
c/o L.I.T.A. - room 25/5
Via Fratelli Cervi 93
I-20090 Segrate (MI)
Italy
Tel. +39-02-21717552
Fax +39-02-21717558
email: [log in to unmask]
|