[MINC-users] nu_correct error

[minc-users@bic.mni.mcgill.ca]

Hi Andrew,

We have found an empirical 'fix' to the nu_correct issue (in case you
might be interested).  The nu_correct runs OK on our problem cases if we
tweak the sampling parameter (set -shrink 4.0001).  We also found that
cropping the images (mincreshape) to zspace=0,192 resulted in nu_correct
working.

As for version bugs, the nu_correct worked on all of our non-linux systems
(results listed below).  Therefore the source of the issue might be
numerical/computational.

Anyway, thanks very much for your feedback.


Ernest Lo
NeuroRX Research
Montreal, Canada


Fails:

N3-1.04 on Linux x86
N3-1.09 on Linux x86
N3-1.10 on Linux x86

Succeeds:

N3-1.09 on SGI
N3-1.05 on Mac OS X 10.3.9



> --__--__--
> 
> Message: 1
> Date: Wed, 6 Jul 2005 22:21:20 -0400
> From: Andrew Janke <a.janke@gmail.com>
> Reply-To: Andrew Janke <a.janke@gmail.com>
> To: minc-users@bic.mni.mcgill.ca
> Subject: Re: [MINC-users] nu_correct error
> 
> Ernest, 
> 
> What version of N3 are you running?  There have been a few "issues"
> with some previous versions....  The current version is 1.10 and can
> be found here:
> 
>    http://packages.bic.mni.mcgill.ca/tgz/
> 
> As for documentation on the field format, you have me there.  The only
> one that could comment on this would be John Sled I guess. (the
> authour).  In the meantime the best way would be to look at the code
> of evaluate_field.  It takes a field file (.imp) and can turn it into
> a MINC file.
> 
> Andrew
> 
> 
> On 06/07/05, elo@neurorx.com <elo@neurorx.com> wrote:
> > Hello,
> > 
> > I have encountered the following error while running nu_correct:
> > 
> > inputCompactField(): Incorrect number of coefficients
> > Failure reading field file:
> > XXXXXXX.imp
> > nu_evaluate: crashed while running evaluate_field (termination status=256)
> > nu_correct: crashed while running nu_evaluate (termination status=256)
> > nu_correct failed on XXXXXXXXX.mnc.gz
> > 
> > Examination of the .imp file shows that it has 90 lines instead of the
> > usual 110 lines.  Therefore it may be that nu_evaluate expects a greater
> > number of field coefficients.  As a test I manually add 20 coefficients to
> > the .imp file and then nu_evaluate works fine.
> > 
> > What could be causing this error in the automated processing?  Does
> > nu_evaluate expect a different number of field coefficients than is
> > provided by nu_estimate?  nu_correct has worked fine on our other scans
> > so this seems to be an uncommon type of failure.
> > 
> > PS it would be very enlightening also to know what the parameters, matrix,
> > and coefficients in the .imp file mean.  Somehow they must represent the
> > non-uniformity field - but what would the mathematical formulation be?
> > 
> > Thanks in advance,
> > 
> > Ernest Lo
> > NeuroRX Research
> > Montreal, Canada