[MINC-users] How to use rawtominc with multiple files

[Ernest Lo]

Some additional comments on N3:

There are arguments for using N3 (nu_correct) before and after
registration.


Reasons to use N3 (nu_correct) before registration:

1. As mentioned by Dr. Narayanan, nu_correction on images may improve
registration.  Having processed a steady stream of multi-modal scans on a
daily basis for almost a year now, I have seen a few instances where
severe image non-uniformity caused poor registration results using
mritoself (the alignment worked fine after nu_correction).  I imagine it
could cause similar effects in mritotal.  Both these algorithms use
intensity-based objective functions as convergence criteria. In any case,
nu_correction should not harm any subsequent registration processes.

2. I have heard that nu_correct works better on isotropic data.  So if one
is resampling anisotropic images to an isotropic space (meaning in this
case that the resolution or step size in the x,y,z directions are all the
same), then this may improve the result.


Reasons to use N3 after registration:

1. Since the nonlinear registration to standard space may both magnify and
reduce areas of the original image, it is possible that small
non-uniformities may be transformed into larger ones by mritotal.  Also,
assuming that N3 runs on the basis of a convergence criterion, this
criterion may no longer be satisfied after nonlinear registration.  
Therefore running N3 after mritotal ensures that all the images satisfy
the same uniformity convergence criterion.


For optimal results, you could run nu_correct both before and after
registration.  In practice, I think the differences will be small
whichever option (before or after) you choose.


Vivek, you mentioned that the default behaviour of nu_estimate is to mask
in Talairach space - are you sure about this?  I detect no nonlinear
transformation processes when nu_correct is run.


Ernest Lo
NeuroRX Research
Montreal, Canada




> ------------------------------
> 
> Message: 8
> Date: Wed, 21 Dec 2005 08:52:13 -0500
> From: Vivek SINGH <vsingh at bic.mni.mcgill.ca>
> Subject: Re: [MINC-users] How to use rawtominc with multiple files
> To: MINC users mailing list <minc-users at bic.mni.mcgill.ca>
> Message-ID:
> 	<Pine.SGI.4.44.0512210849070.85857-100000 at shadow.bic.mni.mcgill.ca>
> Content-Type: TEXT/PLAIN; charset=US-ASCII
> 
> On Tue, 20 Dec 2005, Sridar Narayanan wrote:
> 
> > >> I was trying to resample the input image with average_305.mnc in mritotal
> > >> and mincresample command before using NU_CORRECT. Which would be the better
> > >> solution for NU_CORRECT? directly use on input image or after transforming
> > >> data in a model (average_305) and then use the intensity correction? Please
> > >> give me suggestion in this regard.
> > >
> > > Given that N3 has no knowledge of model space and the likes and works
> > > on the raw histogram I would not expect the results to be different
> > > beyond std interpolation and rounding errors.  Typically N3 is run on
> > > the native data.
> 
> This is not exactly true.  The default behaviour of the nu_estimate script
> (called by nu_correct) is to mask volumes in Talairach space using the
> average_305_mask_1mm.mnc.  (This is done by attaching an -auto_mask option
> when nu_estimate calls the nu_estimate_np_and_em script, for those who
> care)  The decision is based on whether the 'spacetype' variables for the
> x,y and z dimensions of the minc volume are set to "talairach" or not,
> which is done for volumes resampled to an ICBM template using
> mincresample.
> 
> >
> > That's the usual practice, and performing the NU correction first will
> > likely improve the registration to the average brain.
> >
> 
> I believe someone investigated this and found there was little appreciable
> difference in registration using mritotal if the NU correction is
> performed before or after the fact (given data acquired with the ICBM
> protocol), thought it probably doesn't hurt.
> 
> In my experience, performing the NU correction after registration can
> sometimes improve the result, if I am unhappy with the correction before
> registration.
> 
> Vivek