[Insight-users] Perplexed with Optimizer initialization

Luis Ibanez luis.ibanez at kitware.com
Wed May 25 20:45:28 EDT 2005


Hi Yu,

Please post the code that you are using for providing the
initial transform to the RegistrationMethod class.

>From your "#IND"s, it seems that you are not initializing the
last three components of the Parameters array that passes
the initial values to the optimizer.


Please let us know,


    Thanks


        Luis



--------------------------
Yu Qi wrote:

> Hi Luis,
> Thanks a lot!
> Now, I have some "control" over the registration with Mattes algorithm.
> Then I want to register the same images using Viola's method,
> but the the result is not right.
>  
> Also, I use VersorRigid3DTransformOptimizer
> and I set rotationScale=1.0
>              translationScale=1/300
>    maximumStepLength=0.01;
>    minimumStepLength=0.0001;
> and for the metric
>     fixedImageStandardDeviation=0.4;
>     movingImageStandardDeviation=0.4;
>     numberOfSpatialsamples=50;
> But I alway get a error result
> like this
> 0 0.194797  0   0   0   -1.#IND   -1.#IND   -1.#IND  
> 1 0.182961  0   0   0   -1.#IND   -1.#IND   -1.#IND 
> 2 0.23448    0   0   0   -1.#IND   -1.#IND   -1.#IND 
> ...
>  
> my fixed image is 64X64X28 epi
>       moving image is the same one with a translation
>  
> Best regards,
> Yu Qi
>  
>  
> *//*
> */Luis Ibanez <luis.ibanez at kitware.com>/* 写道:
> 
> 
> 
>     Hi Yu,
> 
> 
>     You are right,
>     parameters selection is a problem that is undersestimated by the
>     entire medical image community.
> 
>     This disregards is closely related to the insufficiency of the
>     publishing system and to the lack of appreciation for the practice
>     of reproducibility of research work.
> 
>     In other words: since most published algorithms are not presented
>     in a form that is intended to make easier for other to reimplement
>     and run those algorithms, authors do not address the very important
>     issue of parameter selection. Authors are more interested in claiming
>     credit for the basic ideas behind their published algorithms.
> 
>     In the real world, however, parameter selection is a challenge of a
>     larger scale than software development itself.
> 
> 
>     That being said....
> 
> 
> 
>     Of, course, if you take an example, for w! hich the paramters were
>     already adjusted... and you *change* the parameters... well... it
>     will not always work.
> 
> 
>     Since your question is quite generic,... we can only give you
>     a generic answer. If you are interested in a more specific
>     answer.... please post a more specific question.
> 
> 
>     Generic answer:
> 
> 
>     In the selection of parameters for registration you wan to
>     do the following:
> 
> 
>     1) Make sure that the scaling array of parameter of the
>     VersorRigid3D transform are in the range of
> 
>     1.0 for rotations
>     1/(diagonal of the image in millimeters) for translation
> 
> 
>     2) Connect a command observer and print out the values
>     of the metric and the paramters array at every iteration
> 
> 
> 
> 
>     3) Start with very small steps in the optimizer.
>     E.g. in the range of 0.1
> 
> 
> 
>     Note that Algorithms *do not* and *do not have to* work for
>     *every* arbitrary combination of parameters. So, you shouldnt
>     expect to ! take a fine tunned example and be able to move the
>     parameters at will while still getting reasonable results.
> 
> 
>     Otherwise...
>     well... there will not be any need for parameter fine tunning.
> 
> 
> 
> 
> 
>     Regards,
> 
> 
> 
>     Luis
> 
> 
> 
>     --------------------
>     Yu Qi wrote:
>      > Hi everybody,
>      > I get to know ITK since only few months ago. I find it is a very
>     well
>      > orgnized toolkit for image registration and I made my only
>     application
>      > successfully!
>      >
>      > But there is one thing always confuses me during my programming:
>      > the selection of parameters :-( . I first copy codes from user
>     guide and
>      > it worked. Then I want to change those parameters to see the
>     influence
>      > of them to the results. But it not alway turned out a success.
>      >
>      > Does anyone can help me with how to configure those parameters,
>      > optimizer and metric? What is the most important one which I
>     should take
>      >! ; care first? What is the usual range of adjustment?
>      >
>      > Or will somebody enclose a paper about optimizers to me?
>      >
>      > Metric in my application: MattesMutualInformationImageToImageMetric
>      > MutualInformationImageToImageMetric
>      > Optimizer I used: itkVersorRigid3DTransformOptimizer
>      >
>      > Thank you all!
>      >
>      >
>      >
>      >
>      >
>      >
>      >
>      > *Best Regards!*
>      > **
>      > *Yu Qi*
>      >
>      >
>      >
>      >
>      >
>     ------------------------------------------------------------------------
>      > *Do You Yahoo!?*
>      > 150万曲MP3疯狂搜,带您闯入音乐殿堂
>      > 美女明星应有尽有,搜遍美图、艳图和酷图
>      > 1G就是1000兆,雅虎电邮自助扩容!
>      >
>      >
>      >
>      >
>      >
>     ------------------------------------------------------------------------
>      >
>     &! gt; _______________________________________________
>      > Insight-users mailing list
>      > Insight-users at itk.org
>      > http://www.itk.org/mailman/listinfo/insight-users
> 
> 
> 
> 
> ------------------------------------------------------------------------
> *Do You Yahoo!?*
> 150万曲MP3疯狂搜,带您闯入音乐殿堂 <http://music.yisou.com>
> 美女明星应有尽有,搜遍美图、艳图和酷图 <http://image.yisou.com>
> 1G就是1000兆,雅虎电邮自助扩容! 
> <http://cn.rd.yahoo.com/mail_cn/tag/1g/*http://cn.mail.yahoo.com/event/mail_1g/> 
> 





More information about the Insight-users mailing list