[MINC-users] nu_correct & masking

Fri Jul 29 15:34:04 2005

Hello,

Some further thoughts on the nu_correct issue.

nu_correct works by sharpening the intensity histogram (by successively
deconvolving it).  Therefore, if you mask out the csf, it will determine
the optimal bias correction for the brain-without-csf intensity histogram.  
If masking is not done, then the optimal bias correction for the entire
brain is done.

The 'actual' bias field extends over the entire brain which includes the
csf.  But for processing purposes, the goal may not be to calculate the
'actual' field but rather the one that produces the best uniformity in a
specific tissue type in the resulting image.  So the most 'effective' use
of nu_correct depends on your objective (which would be most likely the
latter, for image processing purposes - hence use a mask).

According to the Sled 1998 article (p.90), "simple thresholding is used to
remove empty regions" from the brain volume.  This may automatically
exclude csf regions as well.

Empirical testing would be useful to verify these ideas.

Ernest Lo
NeuroRX Research
Montreal, Canada

> Message: 1
> Date: Wed, 27 Jul 2005 18:04:33 -0400
> From: Alex ZIJDENBOS <alex@bic.mni.mcgill.ca>
> Subject: Re: [MINC-users] nu_correct & masking
> To: Andrew Janke <a.janke@gmail.com>
> Cc: minc-users@bic.mni.mcgill.ca
> Reply-To: minc-users@bic.mni.mcgill.ca
> 
> Just to chip in: AFAIK it does hurt N3 if you give it data it should
> not optimize over. In other words, assuming you care about the brain
> only, I believe it does help to provide a brain mask such that it
> doesn't try to uniformize scalp/neck/etc.
> 
> -- A
> 
> On Wed, Jul 27, 2005 at 12:26:16AM -0400, Andrew Janke wrote:
> > Hi Richard,
> > 
> > I had hoped that John Sled would bite on this one, but it appears not
> > so here's my thoughts...
> > 
> > My feeling is that (2) is better.  The more data you give N3 to smooth
> > over the better it is going to perform. As far as I am aware, N3 does
> > not optimise over peaks per-se but rather the histogram in general. 
> > Thus as you way, the more "information" in the image the better.
> > 
> > Myself I tend to run nu_correct without a mask altogether beyond any
> > masking it does internally.
> > 
> > Andrew
> > 
> > 
> > On 12/07/05, Richard Boyes <rboyes@dementia.ion.ucl.ac.uk> wrote:
> > > Hello everybody
> > > 
> > > Regarding use of a mask with nu_correct, I have available
> > > to me two types:
> > > 
> > > 1. A brain mask that excludes CSF completely - ie WM, GM
> > >     only
> > > 
> > > 2. A 'filled' brain mask - ie includes the CSF in the sulcal
> > >     and ventricular spaces.
> > > 
> > > My question is, which mask is more effective? Intuitively
> > > I thought that the CSF mask would be better, as it gives 3
> > > classes in the histogram, which gives N3 3 histogram peaks
> > > to optimize over.
> > > 
> > > Thanks in advance
> > > Richard
> > > _______________________________________________
> > > MINC-users@bic.mni.mcgill.ca
> > > http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
> > > 
> > 
> > 
> > -- 
> > Andrew Janke      (a.janke@gmail.com || www.cmr.uq.edu.au/~rotor)
> > Canada->Montreal                                   Cell: +1 (514) 924 2012
> > 
> > _______________________________________________
> > MINC-users@bic.mni.mcgill.ca
> > http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
> 
> --__--__--
> 
> Message: 2
> Date: Wed, 27 Jul 2005 23:09:16 -0400
> From: "Najmeh Khalili M." <najma@bic.mni.mcgill.ca>
> To: Andrew Janke <a.janke@gmail.com>
> cc: minc-users@bic.mni.mcgill.ca
> Subject: Re: [MINC-users] nu_correct & masking
> Reply-To: minc-users@bic.mni.mcgill.ca
> 
> >
> > My feeling is that (2) is better.  The more data you give N3 to smooth
> 
> Some time ago, I tested the effect of performing NU-correct with and
> without a brain mask, and as far as I could tell (and since I was dealing
> with pahological tissue, I "could" tell) masking improved the image
> processing stages that followed N3! So I would suggest option 2 as well.
> 
> Najma
> 
> > >
> > > 2. A 'filled' brain mask - ie includes the CSF in the sulcal
> > >     and ventricular spaces.
> > >
> > > My question is, which mask is more effective? Intuitively
> > > I thought that the CSF mask would be better, as it gives 3
> > > classes in the histogram, which gives N3 3 histogram peaks
> > > to optimize over.
> > >
> > > Thanks in advance
> > > Richard
> > > _______________________________________________
> > > MINC-users@bic.mni.mcgill.ca
> > > http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
> > >
> >
> >
> > --
> > Andrew Janke      (a.janke@gmail.com || www.cmr.uq.edu.au/~rotor)
> > Canada->Montreal                                   Cell: +1 (514) 924 2012
> >
> > _______________________________________________
> > MINC-users@bic.mni.mcgill.ca
> > http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
> >
> 
> 
> 
> 
> --__--__--