[MINC-users] proper use of nu_correct

[Vivek SINGH]

Hi Jon,

Remember that N3 expects the RF imhomogeneity to be multiplicative
and slowly varying across the image.  For example, a 60% imhomogeneity
field (fairly high) has values ranging from about 0.5 to 1.3.  In your
case, the values exceed this variation within a couple of slices.  Maybe
trying a bias field varying within a 0.5 to 1.5 range from the top to the
bottom of the image might be a better test.

Hope that helps,
Vivek

On Sun, 12 Sep 2004, Jon Frederick wrote:

> Hello,
> We're hoping someone can explain what we're doing wrong with nu_correct.
>
> We introduced an artificial bias field into a 256x256x256 MRI image, that
> increased by one (on a pixel intensity scale of zero to 256) with each
> coronal slice (only in the masked area of the brain... the unmasked area
> remained black or zero).
>
> We ran
> nu_correct -iterations 150 -stop .00001 temp.mnc -clobber temp-corr.mnc
>
> After converting the output image back to raw and putting our header back
> on, the bias field could still easily be seen in the sagittal or axial
> views.
>
> nu_estimate of the uncorrected file returns c.v of field change=7.6 e-5.
> For the corrected file, the value is 2.5 e-5. Are both of these numbers
> rather small? John Sled's dissertation shows graphs with c.v.'s on the
> scale of 0.01.
>
> Could the problem be related to our conversion to minc format? I.e.,
> dd conv=swab if=temp.raw | rawtominc -coronal -byte -unsigned -ystep 1
> -zstep 1 -xstep 1 -mri -origin 0 0 0 -range 0 255 -real_range 0 255
> -clobber temp.mnc 256 256 256
>
> Apologies in advance if this is a really dumb question.
>
> Thanks in advance,
>
> Jon Frederick, Ph.D.
> University of Texas Health Science Center
> _______________________________________________
> MINC-users@bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>