[Insight-users] Re: FW: Problem with Demons Variant Deformable Registration

[Torsten Rohlfing]

Hi Sajendra --

There are some known problems with the fast symmetric demons forces code 
currently in ITK. These have been discussed on the mailing list, but 
were never resolved.

Here's an idea though: Tom Vercauteren of INRIA has provided a vastly 
improved variant of the demons algorithm, including among other features 
also a working fast symmetric forces implementation. I believe his code 
is currently in the Review/ directory of the ITK source tree. In any 
case, you can also get his original implementation from the 
InsightJournal website:

 http://insight-journal.org/InsightJournalManager/view_publication.php?back=search.php%3Ftexte%3Dvercauteren&pubid=154

My suggestion would be for you to give Tom's code a try. Maybe that will 
solve your problems.

Best,
  Torsten

Nithiananthan, Sajendra wrote:
> Hello,
>
> Please find below a message I posted to the ITK users group.
> As you two are the authors of the respective classes I was hoping you might
> have some insight.
>
> Thank you for any help you might be able to provide,
> Sajendra N.
>
> -----Original Message-----
> From: Nithiananthan, Sajendra 
> Sent: Wednesday, February 27, 2008 4:56 PM
> To: 'insight-users at itk.org'
> Subject: Problem with Demons Variant Deformable Registration
>
>
> Hello everyone,
>
> I am trying to deformably register two images using the Demons algorithm.
> Based upon the examples in the ITK User's Guide I attempted the registration
> using Thirions original implementation and had good results after a
> sufficient number of iterations. I then attempted the same registration
> using the SymmetricForcesDemons implementation provided with ITK.  I found
> that I do get convergence in fewer iterations, however, each iteration takes
> much much longer to run (100-1000xlonger or worse) on my machine than the
> original implementation.
> After this, I created a Demons Registration filter which used
> itkFastSymmetricForcesDemonsRegistrationFunction (a reimplementation of
> SymmetricForcesDemonsFunction, but more efficient) . This runs at speeds
> comparable to the original Demons implementation, and also converges in
> fewer iterations, similar to SymmetricForcesDemons.  Unfortunately, the
> image I get by using the deformation field produced from
> FastSymmetricForceDemons suffers from degraded image quality: there is a
> "chunkiness" in parts of the image (please see attached image).  This effect
> noticeably reduces the quality of the image to the human eye, but does not
> seem to have too much of a negative impact on the metric the algorithm is
> calculating. This effect does not happen with the original Demons
> implementation or SymmetricForces implementation.
> I have examined the code for all three implementations and can not find the
> source of the problem.  From the code, it is also not clear to me why
> itkSymmetricForcesDemons runs so painfully slow compared to the other
> implementations (it does however only run on one core compared to four for
> the others).
> Would anyone have any insights or suggestions?  The fast convergence of the
> two SymmetricForces algorithms are great, but the incredibly slow running
> speeds and image quality problems are a bit of a show stopper right now.
> If it helps, I can send my sample data and code which produces the
> registered images with degraded quality.
> I got these results using InsightToolkit-3.2.0 compiled with MS Visual
> Studio 8 on Windows XP 32bit.
>
> Thank you very much,
> Sajendra N.
>
>
> This e-mail may contain confidential and/or privileged information for the sole use of the intended recipient. Any review or distribution by anyone other than the person for whom it was originally intended is strictly prohibited. If you have received this e-mail in error, please contact the sender and delete all copies. Opinions, conclusions or other information contained in this e-mail may not be that of the organization.
>   


-- 
Torsten Rohlfing, PhD          SRI International, Neuroscience Program
 Research Scientist             333 Ravenswood Ave, Menlo Park, CA 94025
  Phone: ++1 (650) 859-3379      Fax: ++1 (650) 859-2743
   torsten at synapse.sri.com        http://www.stanford.edu/~rohlfing/

     "Though this be madness, yet there is a method in't"