Hi - I'm afraid it's not very easy for us to help out here, as nreg
isn't our software and, although it works really well on the data
we've put through it, we've not had a lot of experience in tuning it
for problematic data. The good news (depending on your timescale....)
is that the next version of FSL will use "FNIRT" instead, which we've
recently developed in-house, which appears _very_ accurate and stable,
and which we will be able to offer better support for. The timescale
we are _aiming for_ for FSL 4.1 is before the end of May. If this is
too long for you to wait, hopefully you will be able to work out how
to optimise nreg.
Cheers, Steve.
On 8 Apr 2008, at 19:06, Xiujuan Geng wrote:
> Currently I'm using nreg to do FA image registration. I followed the
> parameters
> used in TBSS, specifically, in the script "tbss_reg". For example, I
> followed
> the command as:
> ${D}/nreg ${REFERENCE}.hdr ${INPUT}.hdr -dofin ${I2R}_affine.dof -
> dofout
> ${I2R}_nonlinear.dof -ds 20 -Tp 0 -parameter ${D}/nonlinearCC.par
> But for some data sets, I can't get good registration results.
>
> To figure out in which case nreg works well, and in which case it
> doesn't,
> I simulated two data sets with image size 64x64x64. Please see the
> attachment, in the zip file, there are 3 tiff snapshots from Analyze
> (.hdr/img)
> images due to maximum size limitation: sim1.tif, sim2.tif and
> sime2_to_1.tif.
> The reference is sim1, and the input is sim2. After nreg, I used
> "transform" to
> get the registered sim2_to_1. The registered image doesn't look
> quite similar
> to the reference, especially the big ellipsoid with lower intensity.
>
> I tried the following things:
> 1. increase control points from 20 to 40, i.e., change "-ds 20" to "-
> ds 40";
> 2. increase iteration number inside the parameter, i.e., change "No.
> of
> iterations = 20" to "No. of iterations = 100" in file
> "nonlinearCC.par";
> 3. increase number of bins, i.e., change "No. of bins = 64" to "No.
> of bins =
> 200";
> 4. change the parameter file from nonlinearCC.par to
> "nonlinearSSD.par";
> 5. do linear registration using areg, and then do nonlinear using
> nreg.
> But none of above improved the registration.
>
> Could you please give some hint or suggestion about how to modify the
> parameters to get better registration in this case? Or if I did
> something, could
> you please point it out? If you need the original simulation data,
> please let me
> know, I'll try to send to you. Thanks a lot!
>
> Xiujuan
> <sim_data_1_slice.zip>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
