Dear SPM experts,
I have a question about some resting-state fMRI analyses I am reading
about, which I do not understand.
In these resting-state fMRI analyses, the researchers performed a
seed-to-voxel connectivity analyses by averaging the signal extracted in
a seed ROI per subject during their resting-state scans. This average
signal was then included as the main regressor of interest in 1st-level
models. They performed a one-sample T contrast against zero on this
regressor, then z-transformed the resulting spmT*.nii images, which they
interpret as the connectivity between the seed and the respective voxel
took to their 2nd-level models.
This method strikes me as very odd, perhaps because I just don't
understand the fundamentals and am only more familiar with task-based
fMRI analyses -- but I would have thought that regardless whether it's
rs-fMRI or task-based fMRI, both should still follow some of the basic
statistical applications of these classical level-1 outputs from SPM? I
would have thought that the contrast images reflect the parameter
estimates, arguably here, the signed magnitude of the connectivity
between the seed and voxel, and hence should be what be used in
second-level models if the idea is to test whether the connectivity
changes between conditions/groups -- wouldn't performing any subsequent
tests on the spmT images be like a meta-level statistical test (testing
the test) i.e. testing whether the "connectivity" can be reliably tested
in the way adopted rather than the effect itself? And I don't understand
why you would take the z-transform of the spmT image either -- isn't the
spmT image the studentized T so it's already standardized?
Many thanks in advance for any clarification!
Gina
|