Dear Dr. Friston,
Dear Dr. Ashburner,
Dear SPM experts!
unfortunately we did not receive an answer to our recent questions
regarding
some surprising results we obtained in our second level analysis.
Here we pose these old questions again:
We are analyzing data from a bloc design fMRI study (2 condtitions A and
B, one control condition C, all appearing within one run, two different
groups of subjects 1 and 2, two runs / subject).
We calculated contrast images for each individual: A-C, B-C, A-B
We want to perform group comparisons on the second level. We are
interested in the contrasts:
within group (one sample t-test):
A-C
B-C
A-B
between groups (two sample t-test):
(A1-C1)>(A2-C2),
(B1-C1)>(B2-C2),
(A1-B1)>(A2-B2).
All of the contrast images on the first level look o.k. However some of
the contrasts on the second level (for within [one sample t-test] and
between [two sample t-test] group comparison), in particular those where
we expect very little activation, show significant activation in white
matter areas and areas that are not activated in first level analysis
(in spite of very good realignment parameters, we attribute this
activation to residual head motion).
We did not perform global calculation on the second level, no global
normalisation was performed on the first level.
our questions are:
1 could it be, that if very little variance is present in grey
matter, the significance of signal fluctuation in white matter
increases?
2 is it necessary to perform 'threshold masking' for between
group comparison? (and if yes how is this done? proportional/absolute?
threshold?)
3 for contrast (A1-B1)>(A2-B2) would it be appropriate to include the
control conditions C1/C2 by performing second level analysis of 4
contrast images [(A1-C1)-(B1-C1)]>[(A2-C2)-(B2-C2)]?
4 we also would like to understand, what global calculation on the
second level does.
Thanks very much in advance
Peter Erhard
Bernhard Haslinger
------------------------------------------------------------------
Peter Erhard, Ph.D.
Neurozentrum Funktionelle Bildgebung
TU Muenchen
[log in to unmask]
|