[Insight-developers] SimpleITK: N4MRIBiasFieldCorrection

[Nicholas Tustison]

Hi Brad,

For SimpleITK, I think those are great suggestions.  I agree that returning only 
the bias corrected image would be sufficient and I would also assign the number 
of fitting levels based on the number of iterations so that you can remove the
option 'numberOfFittingLevels' from the command line.  And I would also agree 
that specifying the optional weight image would complicate things unnecessarily 
for the SimpleITK user.  Just specifying a binary mask should be sufficient for 
most cases.

Also, one thing that is crucial to N4's performance that you might want to keep in
mind, at least for brain images, is that the "spline distance"  or number of control
points really affects performance.  Typically, we would set the initial spline distance to
something like 200 mm and then have the number of iterations be something like
50x50x50x50.  So you would have 50 iterations at 200 mm, 50 iters. at 100 mm, 50 iters. at
50 mm, and 50 iters. at 25mm.  If you specify the B-spline mesh using the number
of control points, you might want to start with 4 control points in each dimension and 
then have the number of iterations default to 50x50x50x50.  A single level of 50 iterations
is probably too weak to have much of an effect. 

Nick




On May 13, 2011, at 10:18 AM, Bradley Lowekamp wrote:

> Hello Nick,
> 
> I have an interface of your N4 filter implemented in SimpleITK along with a little example:
> 
> https://github.com/blowekamp/SimpleITK/compare/master...N4BMRIBiasFieldCorrection
> 
> I am thinking that for simpleITK it is sufficient to just return the corrected image, and not deal with the BSpline parameters. Additionally I am debating of all fitting levels should have the sample number of maximum iterators, or the should should set the MaximumNumberOfIterations as an array who's length determines the number of iterations. I am uncertain I am going to incorporate the optional weight image. I am thinking that 2 image will be image, mask, where as there will be image, mask, weight... but I am not sure if the is limiting or simplifying limiting.
> 
> Do you have any comment or suggestions with the current interface?
> 
> Thanks,
> Brad
> 
> ========================================================
> Bradley Lowekamp  
> Lockheed Martin Contractor for
> Office of High Performance Computing and Communications
> National Library of Medicine 
> blowekamp at mail.nih.gov
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/mailman/private/insight-developers/attachments/20110513/04160fc5/attachment.htm>