Greetings from Bethesda:
How can we align sub-millimeter resolution MRI image volumes with FLIRT ?
I am testing flirt with some high-resolution T1-weighted brain images
acquired with a surface coil. The voxel dimensions typically in the range
of 0.4mm*0.4mm*0.6mm to 0.25mm*0.25mm*0.4mm.
I tried using flirt with the following command line:
flirt -in name_20021020_379_5 -ref name_20021020_379_4 -out \
name_20021020_379_5_align_4 -omat name_20021020_379_5_align_4 \
-minsampling 0.1 -finesearch 5 -interp sinc -dof 7
(The subjects are in a bite-bar and vacuum sealed so hopefully they are
not moving very far between sequential T1 images).
The results of this alignmemnt were not spectacular - it looked as
though the mutli-scale estimates of the 7-parameter transform stopped at
1mm (ignoring the 0.1mm minsampling I specified).
After inspecting the source code and noting the tolerance parameters
written there I tried again with the following command line:
flirt -in name_20021020_379_5 -ref name_20021020_379_4 -out \
name_20021020_379_5_align_4 -omat \
name_20021020_379_5_align_4 -minsampling 0.2 -finesearch 5 \
-interp sinc -dof 6 -verbose 3 -noresample
(i.e. no resampling, translate and rotate only).
This seems to be running much slower (after 30 minutes it hadn't
even finished the 8mm scale), so I am running the last command line this
time without the -noresample. This time it converges again more quicjly
but it terminates at the 1mm scale.
No doubt I am doing something foolish but whatever it is it sure beats
me....
Thanks for your help
--
Sean Marrett , Ph.D email: [log in to unmask]
Functional MRI Facility, NIMH/NIH
Room 1D80a, 10 Center Drive, Bethesda, MD. 20892-1148
tel: 301-402-1378 fax: 301-402-1370
|