Ged,
Just a quick comment...
> How does Tom Nichols's comment
> (http://www.sph.umich.edu/~nichols/NIpub/) "Thresholded statistic
> maps can be seriously misleading" relate to all this??
No it doesn't. My comments there concern *statistic* maps, t/F/Z maps
that come out of an analysis. The thresholding under discussion
regards what goes *into* an analysis.
My $0.02 on this (for VBM, fMRI & PET, actually) is that, ideally, a
brain-mask should be used that is based on morphometric operations on
a reference atlas anatomical image (e.g. a mask derived from the output
of descalping program; or see (though this is getting out of date)
http://www-psych.stanford.edu/~kalina/SPM99/Tools/glm_specmask.html ).
However, since you don't always have full brain coverage, such masking
may allow in low-probability (low intensity) voxels to be included;
you don't want these, since their low-prob (low intensity) may be more
a statement about proximity to the edge of the acquired data volume
than actual value in the data. Hence, an 'analysis threshold' may be
desirous. If the data values have an absolute value
(e.g. probabilities) then an absolute analysis threshold can be used,
otherwise a proportional analysis threshold is probably best.
Hopes this helps more than confuses...
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|