At 10:52 25.09.01 +0100, Bryan Strange wrote:
>Dear Morgan,
>
>If you are starting a group analysis from scratch then yes, smooth the
>original images to the desired 'total smoothing' for the reasons outlined
>by Darren. If, however, you have undersmoothed in the first instance,
>there is the facility to smooth the con images before the RFX analysis.
>
>Regards,
>Bryan
Hi Bryan, Darren + all others,
with a bit of delay, just one more thought related to this issue:
It seems to me that there is a specific case where it is actually
advantagous to have two steps of smoothing (i.e. successively smoothing
each subject's functional data [first-level] and the con-images
[second-level]) instead of smoothing the original images to the desired
total smoothness straight away. Two separate steps of smoothing seem
appropriate if you want to maintain as much regional specificity as
possible on the single-subject level. This is the case, for example, for
those types of analyses of functional/effective connectivity that require
the selection of *subject-specific* seed voxels and whose results (e.g.
path coefficients) enter a separate (i.e. SPM-independent) second-level
analysis *without* reference to 3D coordinates (e.g. descriptive or
inferential group-level statistics of path coefficients of a given
connection in order to describe variance or test for specific effects,
respectively, concerning the distribution of these path coefficients across
subjects). For this specific case, it thus seems preferable to smooth the
subjects' functional data with a small kernel X (to make possible the use
of Gaussian random field theory) and to smooth the con-images with a larger
kernel Y (to account for inter-individual differences in brain anatomy),
such that the desired total amount of smoothness is given by sqr(X^2 + Y^2).
In all other cases, I agree with you and Darren that it seems best to
achieve the desired degree of total smoothing at the level of the
individual subject's functional data in one step.
Hope this makes sense - please let me know if it doesn't.
Best wishes,
Klaas
|