Hi Rich,
replies interspersed below, thanks for helping me sort this out.
On Tue, 24 May 2005 15:46:37 -0400
Rich Hammett <[log in to unmask]> wrote:
Juliana Pare-Blagoev wrote:
Hi All,
Thanks for that useful discussion about the left-right flipping issue. I have a follow-up
question.
I have data for which I confirmed that the left, posterior, inferior corner had the
lowest values
and were all positive (voxel [1 1 1] was off screen). This means, I think, that my data
are
right handed and therefore the defaults.analyze.flip value should be set to 0. The raw
images I
was loading in and checking for LPI voxel values were .mnc files. I also confirmed that
the
analyzed results had the low value voxels in the LPI corner.
However, after analyzing the data, I am finding language activation bilaterally, but
stronger on
the right (i.e. on the right in the glass brain). The subjects are children, and the task
was
somewhat difficult for them so it is remotely possible, though not likely, that these
results are
accurate. More likely, however, I am making some mistake and the images are either not
right
handed or there is some other kind of problem.
If anyone has any suggestions for how I can further test to be sure that my data are
indeed right
handed and/or can spot where I'm making a mistake either in setting the default to 0, or
in
reading the results, that would be much appreciated.
If your original mnc files were created correctly, then your import should
be correct.
Were your mnc files created by the scanner, or were they intermediate files
created from scanner output? If the second, what format was the scanner
output?
I collected my data on a Sonata 3T which produces .ima files (proprietary Siemens format). These
.ima files are "read by a program which fixes a number of irregularities (mosiac images are broken
into proper slices with correct position information, time-series are properly ordered and
labelled). This program saves the cleaned up data in a separate location on the disk array
(/space/sharbot/2/minc/) in the MINC file format (a generic format for storage of
multi-dimensional data).
I used minc_split on my .mnc files with no transformations applied. When I checked these single
volume acquisition .mnc files in SPM2 using Display, the images were left-handed (R,P,I was 1 1
1). These are old data, and two waves of subject collection happened. For the second wave, the
data are already right-handed. To make both sets identical, I did the x direction flip on the
left-handed data using SPM2's option within Display. After this flip, the two sets of data appear
to be identically oriented (i.e. within Display they present identically with L,P,I closest to 1 1
1).
There is obviously room here for some kind of operator error. I did this work on the following
assumptions:
(1) if the raw images (whether .mnc or .ima format) have the l,p,i corner voxel values closest to
1 1 1, then those images are right handed and the SPM2 defaults.analyze.flip should be set to
zero.
(2) after processing (realign, smooth, coregister, normalize), l,p,i corner voxel values are still
closest to 1 1 1, then images are right handed and on the glass brain, analysis results should be
left is left, right is right.
Since I'm 99% certain that my results should show left side dominance for the task, I must be
doing something wrong, but I can't find my error.
Does this info shed light on where I've gone wrong and/or what I can do to troubleshoot things
further?
Thanks!
Juliana
The scanners are very exact about this, so there's no reason for the software
to be confused.
rich
-- \ Rich Hammett http://afni.nimh.nih.gov
/ [log in to unmask] sun dogs fire on the horizon
\ meteor rain stars across the night
/ this moment may be brief
\ but it can be so bright
|