Jesper is the expert here, but I have a few ideas: 1) Use as your reference
file the 2mm MNI template, 2) Crop the FOV of your image so that only skull
and brain are included (no neck, no empty space). See the MNI templates for
an example. If these are not enough, perhaps someone else has additional
ideas.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Jayanth Kolar
Sent: Wednesday, February 25, 2009 11:32 AM
To: [log in to unmask]
Subject: [FSL] FNIRT registration for high resolution images
Hi,
I have recently started working on FNIRT and I am having trouble
registering T1 structural images of higher resolution. The program
works great for medium resolution images (256X256X110) but inspite of
some different permutations like changing the --subsamp=8,8,4,4,2,2 ,
I am not able to configure the settings properly for high resolution
images (512x512x110). I am following the method given in the examples
usage on the FNIRT help website and am using teh reference
MNI152_T1_1mm. Unfortunately the computer I am running the program on,
has limited memory (2GB => executing time ~ 1.5hrs) so I cannot try a
lot of different settings.
Could you from your experience please guide me and provide possible
settings for 512x512x110 images. I have copied the log file from a
trial run which did not work for me. Thank you for your help.
Sincerely,
Jayanth Kolar
# name of reference image
--ref=/usr/local/fsl/data/standard/MNI152_T1_1mm
# name of input image
--in=E15778S26
# name of file containing affine transform
--aff=E15778S26_brain_affine.mat
# name of output file with field coefficients
--cout=E15778S26_nonlinear_transf
# name of file with mask in reference space
--refmask=/usr/local/fsl/data/standard/MNI152_T1_1mm_brain_mask
# If =1, use implicit masking based on value in --ref image. Default =1
--imprefm=1
# If =1, use implicit masking based on value in --in image, Default =1
--impinm=1
# Value to mask out in --ref image. Default =0.0
--imprefval=0
# Value to mask out in --in image. Default =0.0
--impinval=0
# sub-sampling scheme, default 4,2,1,1
--subsamp=8,8,4,4,2,2
# Max # of non-linear iterations, default 5,5,5,5
--miter=5,5,5,5,5,10
# FWHM (in mm) of gaussian smoothing kernel for input volume, default
6,4,2,2
--infwhm=8,6,5,4.5,3,2
# FWHM (in mm) of gaussian smoothing kernel for ref volume, default 4,2,0,0
--reffwhm=8,6,5,4,2,0
# Weight of regularisation, default depending on --ssqlambda and
--regmod switches. See user documetation.
--lambda=300,150,100,50,40,30
# Estimate intensity-mapping if set, deafult 1 (true)
--estint=1,1,1,1,1,0
# (approximate) resolution (in mm) of warp basis in x-, y- and
z-direction, default 10,10,10
--warpres=10,10,10
# Order of spline, 2->Qadratic spline, 3->Cubic spline. Default=3
--splineorder=3
# If set (=1), lambda is weighted by current ssq, default 1
--ssqlambda=1
# Allowed range of Jacobian determinants, default 0.01,100.0
--jacrange=0.01,100
# Model for regularisation of warp-field [membrane_energy
bending_energy], default bending_energy
--regmod=bending_energy
# Model for intensity-mapping [none global_linear global_non_linear
local_linear global_non_linear_with_bias local_non_linear]
--intmod=global_non_linear_with_bias
# Order of poynomial for mapping intensities, default 5
--intorder=5
# Resolution (in mm) of bias-field modelling local intensities, default
50,50,50
--biasres=50,50,50
# Weight of regularisation for bias-field, default 10000
--biaslambda=10000
# Precision for representing Hessian, double or float. Default double
--numprec=double
# If =1, ref image is used to calculate derivatives. Default =0
--refderiv=0
--
Jayanth Kolar
Graduate Research Assistant
Physiological Imaging and Modeling Lab
University of Illinois at Chicago
Chicago IL - 60607
1-408-398-0097
|