[Insight-users] Multi-modal volume registration using mutual
information
Luis Ibanez
luis.ibanez at kitware.com
Mon Jun 28 23:32:25 EDT 2004
Hi Luke,
Are you using Multi-Resolution ?
The way to get registrations working below 2 minutes
require at least two levels in a multi-resolution pyramid.
You don't need to do smoothing before subsampling since it
takes too long and it is not critical for getting the
registration done.
Powell is a relatively primitive optimization method.
We usually work with the GradientDescent optimizer, and in
the case of Mutual-Information it is reasonable to use the
OnePlusOne evolutionary optimizer.
Also, you probably don't want to use an Affine transform for
the two coarsest levels of the pyramid. A rigid transform will
do it ok at those levels, and only has 6 parameters to be
optimized.
Please try the ImageRegistration plugin in VolView.
You can download the free version of VolView from
http://www.kitware.com/products/volview.html
and you will find the source code of the plugins in
InsightApplications/VolviewPlugins
Regards,
Luis
------------
Luke wrote:
> Hi, Luis,
>
> Thank for your reply!
>
> I have read through the tutorial. It is very helpful.
>
> I have a C++ program that register 3-D image using mutual information. You
> claim that I can have reasonable registration in the order of one or two
> minute for 200x200x200 image. However, my program would need a little bit
> more than 1 hour to register 255x255x54 image to a template(on my 1.7GHz,
> 500MB laptop), which is unacceptable.
>
> My program uses powell method(code from numerical recipes in c) for
> optimization. The mutual information is computed using histogram. And the
> transformation matrix is affine transformation with 12 components.
>
> I didn't see itk use powell method as the optimizer. Would other
> optimization method have same accuracy as powell?
>
> How much time would itk take to register a 255x255x54 dataset with 12
> componets affine transformation? Can I find some examples? I want to compare
> before I decide to change my program to itk.
>
> Thank for your help!
>
> Best regards!
>
>
> Luke
>
> ----- Original Message -----
> From: "Luis Ibanez" <luis.ibanez at kitware.com>
> To: "Luke" <khua1 at jhu.edu>
> Cc: <insight-users at itk.org>
> Sent: Thursday, June 24, 2004 6:20 PM
> Subject: Re: [Insight-users] Multi-modal volume registration using mutual
> information
>
>
>
>>Hi Luke,
>>
>>Yes,
>>There is a lot of support for Multi-Modality registration in ITK.
>>
>>There are three different implementations of Mutual Information
>>in ITK. They are:
>>
>> - Viola & Wells
>> - Mattes et al.
>> - Histogram based
>>
>>You will find the full list of Image Metrics in the Groups section
>>of the Doxygen documentation:
>>
>>http://www.itk.org/Insight/Doxygen/html/group__RegistrationMetrics.html
>>
>>
>>Registration is not necesarily time consuming. It is a matter of
>>trade-offs. Usually you can get reasonable registrations in the
>>order of one or two minutes for volumes of size 200x200x200.
>>
>>However, getting there requires that you fine tune the parameters
>>of the componentes involved in the registration process, and of
>>course that you select an appropriate combination of such components.
>>
>>A simple way of checking if you really need registration is to
>>resample one image into the coordinate system of the other image
>>using an identity transform. Then computing the Difference between
>>the resampled image and the fixed image. If you visualize such
>>image you will rapidly notice how much miss-registration exists
>>between the two.
>>
>>You are strongly encouraged to read the Chapter on Image Registration
>>in the Software Guide
>>
>>
>> http://www.itk.org/ItkSoftwareGuide.pdf
>>
>>
>>Chapter 8, pdf-page 241.
>>
>>Mulit-Modality registration is discussed in section 8.4,
>>pdf-page 255. You will find the associated source code
>>examples in the directory:
>>
>>
>> Insight/Examples/Registration
>>
>>
>>
>>You should also read the Tutorial sessions
>>
>> http://www.itk.org/HTML/Tutorials.htm
>>
>>In particular
>>
>>http://www.itk.org/CourseWare/Training/RegistrationMethodsOverview.pdf
>>
>>
>>Please let us know if you have further questions.
>>
>>
>>
>> Thanks
>>
>>
>> Luis
>>
>>
>>-----------
>>Luke wrote:
>>
>>
>>>Is there any multi-modal volume registration using mutual information
>>>inplemented in itk?
>>>
>>>BTW, the registration process is very time consuming, and some times
>>>it turns out the two volume don't need registration at all(no
>>>transformation, rotation, etc). Is there any way I can tell if the two
>>>volume needs registration before optimization?
>>>
>>>Thanks!
>>>
>>>
>>>Luke
>>>
>>>
>>>------------------------------------------------------------------------
>>>
>>>_______________________________________________
>>>Insight-users mailing list
>>>Insight-users at itk.org
>>>http://www.itk.org/mailman/listinfo/insight-users
>>
>>
>>
>
>
More information about the Insight-users
mailing list