Hi Jason,
Which part of the five page Methods section didn't you understand? ;-)
Seriously, though, was there a step or two that seemed particularly
opaque? Here is my readers' digest condensed version:
* Register each structural volume to wustl.edu's 711-2C space via affine
transform (711-2C is based on the ICBM template, something about
Lancaster, I think).
* Segment structural volume using SureFit (now part of Caret);
tessellate segmentation -> midthickness 3D surface.
* Generate cerebral hull volume: Dilate segmentation volume six times
and erode by six times to fill sulci, but keep overall brain size same;
tessellate hull.
* Generate depth (midthickness node-scalar mapping): Find distance from
fiducial surface node to closest cerebral hull node.
* Flatten and register midthickness surface: Use Core6 landmarks to
constrain spherical deformation. (Flattening provides an easy way to
draw registration landmarks.)
* Apply deformation map to depth -> one depth column/file for each
subject all on PALS_B12 standard mesh.
* Average resulting depth columns/files.
Segmentation is by far the most difficult, time-intensive step; it's
downhill from there.
Since the PALS_B12 paper, we have been using "t-maps" to look for
anatomical differences across populations. We hope you will soon be
reading about this in the Journal of Neuroscience, if I can ever
complete some important enhancements to the depth generation algorithm
-- important enough for us to rework all the figures (but not change the
ROIs much).
One important consideration for you is your choice of landmarks. Using
the Core 6 landmarks will normalize away any differences in the central
sulci, because the central sulcus is one of the landmarks. If you want
to align the central sulci, this is good; if you're looking for
cross-group differences there, this is bad. You can delete that
landmark, but keep the others (and perhaps add another elsewhere), but
this is something to think about. Importantly, you can run the
registration both ways, using different deformation prefices (e.g.,
defCore6_ and defNoCeS_), and create average depth and/or t-maps using
the respective results. Each result will be valid in context, but will
tell you something different.
Hope this helps.
On 03/28/2006 06:08 PM, Jason D Connolly wrote:
Dear Caret-users,
Could someone please instruct me as to how the spherical and flattened
maps were averaged in the van essen 05 paper? We hope to create an avg
struct image with the pixel intesity reflecting the degree of
overlap/similarity across anatomical datasets (see figs 2 and 6). The
goal is to see how the central sulci line up across subjects.
Many thanks, Jason.
------------------------------------------------------------
Jason D. Connolly, PhD
Center for Neural Science, New York University
6 Washington Place Room 875, New York, NY 10003
cell:646.417.2937 lab:212.998.8347 fax:212.995.4562
http://www.psych.nyu.edu/curtislab/people/jasonconnolly.html
_______________________________________________
caret-users mailing list
caret-users@brainvis.wustl.edu
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
--
Donna L. Dierker
(Formerly Donna Hanlon; no change in marital status -- see
http://home.att.net/~donna.hanlon for details.)