[Insight-users] Re: Mutual Information and Image Registration

Sun Sep 5 11:02:00 EDT 2004

Hi Carole,

The Joint Entropy is not estimated by the interpolator.
The estimation is done by the ImageMetric itself.

You may want read some of the papers on Mutual Information that
are recommended in the Doxygen documentation of the following classes

http://www.itk.org/Insight/Doxygen/html/classitk_1_1MutualInformationImageToImageMetric.html

http://www.itk.org/Insight/Doxygen/html/classitk_1_1MattesMutualInformationImageToImageMetric.html

For example:

0) Viola, P. and Wells III, W. (1997).
   "Alignment by Maximization of Mutual Information"
    International Journal of Computer Vision, 24(2):137-154

1) Mattes et al.
  "Nonrigid multimodality image registration"
   D. Mattes, D. R. Haynor, H. Vesselle, T. Lewellen and W. Eubank
   Medical Imaging 2001: Image Processing, 2001, pp. 1609-1620.

2)"PET-CT Image Registration in the Chest Using Free-form Deformations"
   D. Mattes, D. R. Haynor, H. Vesselle, T. Lewellen and W. Eubank
   IEEE Transactions in Medical Imaging. Vol.22, No.1, January 2003.
    pp.120-128.

3)"Optimization of Mutual Information for
   MultiResolution Image Registration"
   P. Thevenaz and M. Unser
   IEEE Transactions in Image Processing, 9(12) December 2000.

You will also find a description of Mutual Information in the
book that we recommended in our previous email.

>>
>> Medical Image Registration
>> by Joseph Hajnal, D. J. Hawkes, Derek Hill.
>>
>>             http://www-ipg.umds.ac.uk/d.hill/hhh
>>
>>

If you are looking for new approaches you may want to explore the
notion of "object based information". Please read the discussion
in the Wiki pages

    http://www.itk.org/Wiki/ITK_Mutual_Information

where the notion of "pixel-based information" is discussed.
You will see that the current notion of Entropy in an image is
derived from the assumption that every pixel is independent and
that all the information is contained in the individual pixel
intensities.  This disregards the fact that the real information
in an image is stored in the spatial context of the pixels, e.g.
the actual shapes on the image. A more powerful and comprehensive
use of Mutual Information can be made by using the entropy measured
from OBJECTS in the scene, instead of the primitive entropy measured
from individual pixels.

The use of Mutual Information in image processing was borrowed
(as almost everything else) from other domain. Specifically from
the domain of Communication Theory. Most people got satisfyied with
using individual pixels as symbols and the image as a communication
channel. This classical approach involves a great waste of information,
since it basically states that spatial arrangement of pixels in the
image are totally irrelevant.

An application of Mutual Information based on the knowledge that there
is an anatomical object in the image (e.g. a Brain, a Liver...) instead
of just a random set of pixel values, should provide a higher level
approach to image analysis and to image registration.

   Regards,

      Luis

----------------------
张 卡璐 wrote:

> 
> Hi Luis,
> 
>    Thank you very much for your advice.
>    I need to publish some papers to obtain my doctoral degree. But I 
> just come into this research domain,so I need to find new ideas in 
> medical image registration.
>    As to your "estimation of joint entropy in Mutual Information", my 
> understanding is that happens in interpolator stage, don't it?
>    Maybe i don't understand you, please tell me more.Thanks a lot.
> 
>  
>   Thanks,
>  
>   carole
>   
> 
>> From: Luis Ibanez <luis.ibanez at kitware.com>
>> To: 张 卡璐 <carole_zhang780716 at msn.com>
>> CC: Insight-users at itk.org
>> Subject: Re: i am puzzled,please help me! : Mutual Information and Image 
> 
> Registration
> 
>> Date: Sat, 04 Sep 2004 12:11:08 -0400
>>
>>
>> Hi Carole,
>>
>> ITK provides three different implementations of the Mutual Information
>> metric.  For a full account of the Image Metrics available in ITK
>> please look at
>>
>> http://www.itk.org/Insight/Doxygen/html/group__RegistrationMetrics.html
>>
>> For a detailed explanation of the Image Registration framework in ITK,
>> including the Image Metrics, please read the ITK Software Guide
>>
>>         http://www.itk.org/ItkSoftwareGuide.pdf
>>
>> Chapter 8, pdf-pages 241 to 340.
>>
>>
>> Read also the tutorials on Image Registration methods
>> that are available at
>>
>>     http://www.itk.org/HTML/Tutorials.htm
>>
>> in particular
>>
>> http://www.itk.org/CourseWare/Training/RegistrationMethodsOverview.pdf
>>
>> and
>>
>> http://www.itk.org/CourseWare/Training/NonRigidRegistrationMethods.pdf
>>
>>
>> You will also find useful the online book on Medical Image Registration
>> by Joseph Hajnal, D. J. Hawkes, Derek Hill.
>>
>>             http://www-ipg.umds.ac.uk/d.hill/hhh
>>
>> A forum for discussions in Image Registration is available in the
>> ITK Wiki pages
>>
>>               http://www.itk.org/Wiki/ITK
>>
>> more specifically at
>> http://www.itk.org/Wiki/ITK_Mutual_Information
>>
>>
>> ------------
>>
>> Is your interest in "innovation" motivated by the need of publishing
>> a paper in order to include it in your CV ?
>>
>> If so, you may want to explore the intricate mathematical issues related
>> to the estimation of joint entropy in Mutual Information. That will
>> generate enough equations for making your paper look involved. It may
>> be relatively useless for real applications, but that doesn't really
>> matter if your goal is to publish.
>>
>> If instead of just publishing, you are interested in further developing
>> techniques that may actually be useful for processing images for medical
>> applications, then you probably may want to take a look at more
>> sophisticated Image Metrics that may simultaneously combine Mutual
>> Information with Edge information. This has been proposed in the past,
>> but we don't have an implementation in ITK yet. As you will notice from
>> reading the documents above, Mutual Information focuses in matching
>> large regions on the images. It doesn't particularly consider the edges
>> of structures in the images. It is therefore interesting to introduce
>> some edge-related factors in order to obtain a more powerful metric.
>>
>>
>>
>> Please let us know if you have further questions.
>>
>>
>>    Thanks,
>>
>>
>>       Luis
>>
>>
>>
>> --------------------------
>> 张 卡璐 wrote:
>> >    hi, i am a new user of ITK, and i am engaged in research in medical
>> > image processing, especially image registration. As far as I know,
>> > mutual information is used to image registration and is quite 
> 
> effective.
> 
>> >    I want to combine ITK with mutual information to apply to the 
> 
> medical
> 
>> > image registration. But I am puzzled how to innovate.    Please give
>> > directions to me.
>> >    Thanks for your advice!
>> >
>> >
>> >
>> >  carole
>>
>>
>>
> 
> _________________________________________________________________
> 享用世界上最大的电子邮件系统― MSN Hotmail。  http://www.hotmail.com 
> 