[MINC-users] help on using rawtominc and n3

[Ernest Lo]

> Message: 1
> Date: Wed, 30 Nov 2005 21:59:56 +0000
> From: Runa Parveen <runa at medphys.ucl.ac.uk>
> Subject: [MINC-users] help on using rawtominc and n3
> To: minc-users at bic.mni.mcgill.ca
> Message-ID: <5.2.0.9.0.20051130211322.00bbf3b8 at pop3-server>
> Content-Type: text/plain; charset="us-ascii"; format=flowed
> 


> 4. Also after applying nu_correct, the program runs only for 2 iteration 
> and given CV: 0.000878881, is it OK?
> 
> 5. As it is stopped after 2 iterations, I use the command:
> nu_correct -iterations 100 -stop 2.e-5 data_pix_resample.mnc data_pix_nu.mnc
> 
> it stopped after 37 iterations and CV of field changes: 1.61641e-05.
> 
> Also, in Display, the view of the image in tilting position. Also I would 
> like to know how I could understand the non-uniformity has been corrected?
> 


Hi Runa,

As far as I can tell nu_correct does not provide a way to assess the
degree of non-uniformity of an image.  The CV of field change condition
does not guarantee acceptable uniformity in a single image, or consistency
in uniformity across images.

For our purposes, we find that it is useful to examine the bias field
removed by nu_correct.  You can determine this readily by dividing the nu
corrected image by the original (i.e. mincmath -div original.mnc
corrected.mnc bias_field.mnc).  We run nu_correct repeatedly until the
resultant bias field falls below a threshold variation.

When nu_correction finishes quickly - 2 or 37 iterations is relatively
quick - it could mean that there is very little non-uniformity in the
image, or that the stopping condition is too high.  You need to experiment
with reducing the stopping condition, although I would guess (given -stop
2.e-5)  that your images are quite uniform to begin with.

With regard to your last question, whether or not an image has been
corrected successfully depends on the intended application.  For example,
our goal is to perform automatic tissue classification.  So successful
nu_correction means that the tissue intensity distributions have become
narrow enough to produce acceptable classification results.  A good way to
visually assess image uniformity however is to view the images in
'spectral' - the various tissue types (sulci, white matter, gray matter,
csf) should each have a relatively consistent colour range.

Hope this helps,


Ernest Lo
NeuroRX Research
Montreal, Canada