[Insight-developers] MattesMutualInformation failures in ITK4

[Stephen Aylward]

Typically happens when the parameter scales are set incorrectly.

If the scales are too small then the optimizer may make too large of a
step (step size is about 1/scale) for a particular parameter.   This
could result in an image shooting past its target...so the joint PDFs
will sum to zero (indicating too few pixels overlap).

Scales should be 1/expected_value.   So, if you expect the shift
needed to solve a particular problem is 10mm, then the scale for the
offsets should be 1/10, then a unit step by the optimizer will lead to
the expected value.

s

On Tue, Feb 1, 2011 at 7:28 PM, Johnson, Hans J <hans-johnson at uiowa.edu> wrote:
> By the way, this exact same code works when compiled against ITKv3, and is
> tested extensively and is included as part of the Slicer3 project.
> Hans
>
> From: Kent Williams <norman-k-williams at uiowa.edu>
> Date: Tue, 1 Feb 2011 08:54:50 -0600
> To: ITK <insight-developers at itk.org>
> Cc: Hans Johnson <hans-johnson at uiowa.edu>
> Subject: MattesMutualInformation failures in ITK4
>
> We have an application for Image Registration, BRAINSFit
> (http://www.nitrc.org/projects/multimodereg/) that is used at Iowa for our
> standard workup, and is also distributed with Slicer.
>
>
>
> As part of the Slicer effort and our ongoing work on ITK4 we have begun
> testing BRAINSFit built with ITKV4.  Hans Johnson asked me to look into the
> testing failures with ITKV4.
>
>
>
> What I've discovered is that most of the failures come down to this an
> exception with the error message  "Joint PDF summed to zero" -- entire
> message reproduced below.
>
>
>
> Now, I'm a software process guy more than an image processing guy, and I
> don't pretend to understand very much about Image Registration algorithms or
> the Mattes method in particular, so I'm asking for a little help on this
> one. It doesn't seem right that code that produces useful results in ITK3
> should fail catastrophically in ITK4.
>
>
>
> Any ideas?
>
>
>
> itk::ExceptionObject
> (0x102811a70)
> Location: "void itk::MattesMutualInformationImageToImageMetric<TFixedImage,
> TMovingImage>::GetValueAndDerivative(const typename
> itk::ImageToImageMetric<TFixedImage, TMovingImage\
>>::ParametersType&, typename itk::ImageToImageMetric<TFixedImage,
>> TMovingImage>::MeasureType&, typename itk::ImageToImageMetric<TFixedImage,
>> TMovingImage>::DerivativeType&) cons\
> t [with TFixedImage = itk::Image<float, 3u>, TMovingImage =
> itk::Image<float,
> 3u>]"
> File:
> /scratch/kent/itktest/ITK/Code/Algorithms/itkMattesMutualInformationImageToImageMetric.txx
> Line:
> 1030
> Description: itk::ERROR:
> MattesMutualInformationImageToImageMetric(0x101ccb990): Joint PDF summed to
> zero
>
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://kitware.com/products/protraining.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-developers
>
>



-- 

==============================
Stephen R. Aylward, Ph.D.
Director of Medical Imaging Research
Kitware, Inc. - North Carolina Office
http://www.kitware.com
stephen.aylward (Skype)
(919) 969-6990 x300