Hi Rebecca,

As John said, uploading your volume is the quickest way for us to help, but here are some additional thoughts/factors:

* For very non-uniform volumes, 3dUniformize may not be enough. You might need a bigger gun like FSL's FAST (http://www.fmrib.ox.ac.uk/fsl/fast/index.html), which does better if you skull-strip with BET first. I have a script named "fsl_bias_corr.sh" that takes an input Analyze .hdr and runs BET and FAST on it. You may need to tweak your FIT and BIAS_SMOOTHING_ITERATIONS iterations to improve your result:



INBASENAME=`echo $INFNAME | sed 's/.hdr$//g'`
BETBASENAME=`echo $INFNAME | sed 's/.hdr$/_bet/g'`

echo "========="; echo $COMMAND ; echo "========="

echo "========="; echo $COMMAND ; echo "========="

* After running FAST, you may find the problem disappears. It's likely failing during eye or hindbrain removal, both of which are rather brittle parts of the algorithm. Things that make this part break most often are:

-- rotation off AC-PC line (see AC work-around below, but there are definitely limits to how far off AC-PC the volume can be)
-- little CSF clearance between ventral occipital cortex and cerebellum
-- blood vessels or optic nerve preventing disconnection from skull

* Besides correcting for gain/bias field, other things that can help work around this problem are:

-- peak tweaking: sometimes the difference between WMpeak=75 and WMpeak=76 can be dramatic
-- adjusting the AC position, especially in the Z (axial) dimension

On 12/06/2006 04:48 PM, John Harwell wrote:

Can you upload your volume to http://pulvinar.wustl.edu/cgi-bin/upload.cgi and we will see what is going on. Please let us know when you have uploaded your volume and the name of the volume file.

John Harwell

Department of Anatomy and Neurobiology
Washington University School of Medicine
660 S. Euclid Ave. Box 8108
St. Louis, MO 63110 USA

On Dec 6, 2006, at 2:48 PM, Rebecca Ebitz wrote:

Hi - I'm having a problem segmenting normalized scans.

I've got a very non-uniform volume (macaque, from a 4.7T w/ 0.5 mm voxels) and have used AFNI's 3dUniformize to correct these issues. However, when I try to segment the normalized data in CARET, it either gives me an error (usually that findBiggestObjectWithinMask failed in some way) or it simply crashes.

I can run a successful segmentation on the non-uniformized volume, it just requires a great deal of manual correction. I'm doing the same processing steps for both the normalized and non-uniform volumes.

Anything I should try here? I've tried to run several different normalizations and all of them cause segmentation to fail.

Please let me know what other information you need from me on this one.


Rebecca A. Ebitz, post-bac IRTA
Laboratory of Brain and Cognition
National Institute of Mental Health
Building 10, Room 4C214
Office: 301.4510.8509

caret-users mailing list
caret-users mailing list

Reply via email to