Hi Graham,
Burying your head in David's PALS paper is indeed the best way to get
your head around this stuff. It took me three careful reads before I
really digested it. (And you can't whip through in a short plane ride.)
To get a more hands-on feel for Caret/SureFit surfaces, consider
downloading and playing with these datasets in Caret:
PALS_VE05_FIG-11_FMRI_MFMapping.73730.spec
From PALS paper, figure 11
http://sumsdb.wustl.edu/sums/archivelist.do?archive_id=6334372
Human.PALS.VE-et-al_JN06_WILLIAMS_Suppl_Fig-1_Volumes_Fiducials.spec
From WS paper (http://www.jneurosci.org/cgi/content/full/26/20/5470),
supplementary figure 1
http://sumsdb.wustl.edu/sums/archivelist.do?archive_id=6473226
Also, notice in David's email replies: You have taught him to say
voxel-wise statistics rather than volume-averaged fMRI results --
well done!
When mapping to the PALS_B12 surfaces, there's plenty of uncertainty to
go around. Increasingly, there's consensus that reconstructing
individual surfaces and doing one's analysis in surface-land is an
excellent complement to voxelwise-statistics for most cortical studies.
But for the majority of researchers not yet prepared to do so, this
atlas is a useful visualization substrate for their group results.
Questions about which mapping method to use are valid, and reasonable
people can disagree about such things, but as you say, it's relatively
in the noise compared to what I would call doing it right.
Another source of variability stems from the degree to which the
PALS_B12 average fiducial surface represents your sample's average
anatomy well. In a joint study with Dr. Csernansky and Deanna Barch's
lab, the PALS_B12 average fiducial represents the CONTROLS' anatomy
well, but schizophrenics' anatomy less well. Anatomical differences
across groups almost certainly contribute (even where they may not quite
reach statistical significance by our conservative method). In the
attached captures, the structural underlay is a mean SCZ volume for
SCZ13anat_SFSdiff.jpg, mean CONTROL volume for CON19anat_SFSdiff.jpg.
In both captures, the blue contour is the average CON fiducial (sample
subjects -- not PALS_B12); the red contour is the average SCZ fiducial.
Not shown, the CON average fiducial overlaps the PALS_B12 fiducial
much better than does the SCZ average fiducial. Of course, you don't
know how good a proxy PALS_B12 is for your sample unless you reconstruct
all your subjects. (Thanks to Dr. Lei Wang and Dr. Deanna Barch for
allowing use of their prepublication data.)
On 10/27/2006 12:39 AM, Graham Wideman wrote:
David:
Further, my superficial understanding of MFM worries me that a weak
activation caught by all surfaces might produce the same appearance
as strong activation caught by only a small proportion of the MF
surfaces.
That is indeed possible. However, this isn't necessarily a bad
thing. Moreover, the same thing can occur during volume-based
statistical analyses - a voxel that shows weak but consistent
activation may show the same z-statistic as a voxel that has a strong
activation in a small proportion of individuals.
Not sure if it's important, but these seem somewhat different cases.
In the voxel-based situation the confounding obscures the difference
between a few strong activations vs many weak activations.
In the MFM situation, the confounding is at a subsequent stage: this
is not about muddling strong and weak activations, but rather it's
about the sampling strategy (fMRI-surface mapping) being more or less
sensitive in different regions, depending on the variability of the MF
surfaces in that region (ie: how many of the MF surfaces intersect an
activation).
But all this may be relatively in the noise, I guess, compared to
using surfaces derived from each individual subject.
(Again, time to read the paper in detail, and see in what ways you've
been able to characterize the sensitivity of generic MFM vs using
subjects' own surfaces.).
All very interesting.
Graham
--
Donna L. Dierker
(Formerly Donna Hanlon; no change in marital status -- see
http://home.att.net/~donna.hanlon for details.)
inline: SCZ13anat_SFSdiff.jpginline: CON19anat_SFSdiff.jpg
