Greetings fellow SPM'ers,
I'm currently trying to optimize my VOI extraction process for use in
PPI/DCM. As such, I am trying to extract the time series from FFX SPM's
on unsmoothed, unnormalized data, so as to best account for individual
variability in structural and functional anatomy.
My question is related to the problems with deciding statistical
thresholds before getting spm_regions to extract the data, as discussed
<https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0705&L=SPM&D=0&P=167635>previously
on the list. While I can pretty much understand the
specificity/sensitivity considerations when defining a VOI based on
smoothed, normalized data, I am less certain about the issue when these
two conditions have not been met (as in my case).
The problem I am facing stems from the fact that it is rather difficult
to follow the recoomendations suggested in e.g. this thread
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0705&L=SPM&D=0&P=167635
<https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0705&L=SPM&D=0&P=167635>
of using a somewhat strict uncorrected threshold for selecting the
voxels one wants. I believe D. Gitelman has suggested that this approach
is optimal in relation to PPI's. However, trying to apply this to
unsmoothed data most often leaves me with no activated voxels in the
vicinity of the region of interest, or if there actually are activated
voxels, there are typically <10 voxels included in a VOI with 5 mm radius.
I have tried working around this by setting the threshold to T=1. I am
unsure, however, if this is an optimal solution to my problem, as this
is bound to radically impact on the sensitivity of my PPI, through the
inclusion of a large amount of noisy voxels in my data.
So- if the experts would care to chime in on what, exactly, would be the
optimal approach for working with unsmoothed data, I'd greatly
appreciate it.
Best wishes,
Haakon Engen,
University of Oslo, Norway.
|