Dear Spmrs,
I have been trying to understand the isseues involved in choosing the
size of a smoothing kernel but have a few questions regarding group
analysis.
1) Is the size of the kernel determined by the size and anatomical
variability of a structure. So for example if looking for changes in the
hippocmapus i could smooth at 6/8 as my voxel size is 1x1x1, the structure
does not vary greatly and the overall size of the structure is relatively
small, so a small kernel would smooth any variability to the edges where it
could be deteced as partial volume effects. Then for more complex
structure like cortex i would need to use a larger kernel for example 10,
so thta i'm not getting diffrences that are due to systematic registration
errors. Is this right? Does this mean then that if i am looking in a
subject population for changes in cortex and hippocampus i need to do two
sets of anlysis?
2) Why is the smoothing kernel i input different to the parameters that
are reported when SPM does the analysis, and why does this vary according
to design matrix?
3) Do the same issues regarding structure size and variability apply to
white matter or does the amount of smoothing have to be increased for some
other reason? So again if looking at the white mattter tracks around
heschls gyrus is that a different kernel size to looking at the corpus
callosum. I have some white matter analysis already and smoothed at 10FWHM
two areas appear, one more prominant than the other, but when smoothed at
12, the order of prominance reverses, please can someone explain this.
Thank-you in advance for your advice ...
Emma
|