[Insight-users] Re: Speed Image and Propagation, Advection and Curvature terms

[Luis Ibanez]

Hi Conn,

The output of the Sigmoid is strictly speaking a "feature" image.

 From this image the PDE framework computes a "speed" image. The
speed image is computed as a combination of terms:


       speed = propagation + curvature + advection


the propagation term is = feature image X propagationWeight



For more details you should read the ITK Software Guide

        http://www.itk.org/ItkSoftwareGuide-2.0.1.pdf

In particular Section 9.3, "Level Set Segmentation",
in pdf-pages 496 to 541.


   Regards,


      Luis


--------------------
conn sullivan wrote:

> Hello Luis,
>  
> It is not clear to me yet how speed is computed from Speed Image 
> obtained from Sigmoid?
> Is propagation term directly taken from values of speed image? What 
> about curvature and advection terms?
>  
> Thanks,
> Conn
> 
> */Luis Ibanez <luis.ibanez at kitware.com>/* wrote:
> 
> 
>     Hi Conn,
> 
>     You may want to try:
> 
>     EdgePotential
>     http://www.itk.org/Insight/Doxygen/html/classitk_1_1EdgePotentialImageFilter.html
> 
>     and he ExpNegativeImageFilter
>     http://www.itk.org/Insight/Doxygen/html/classitk_1_1ExpNegativeImageFilter.html
> 
> 
>     The first compute the gradient of the image and then pass it through an
>     exponential. The second only do the exponential. The advangate of the
>     second is that you can select your own gradient-magnitude-like filter.
>     The disadvantage is that you use two filters.
> 
> 
>     Note... however that the reason why we moved into using the Sigmoid for
>     constructing speed images was that the traditional methods (edge
>     potential) do not offer enough flexibility for controlling the creation
>     of the speed image.
> 
> 
>     I would suggest you to reconsider the way you are using the Sigmoid
>     filter...! you probably just need to find better values for Alpha
>     and Beta.
> 
> 
>     Note also that if the gradient magnitude image have gaps in the contours
>     there is not much that you can improve over that when computing the
>     speed image. Probably the only thing that can help you there is to
>     increase the CurvatureScaling of the level set...
> 
>     Presence of gaps in the gradient magnitude simply means that the
>     intensities of the object that you are trying to segment are too
>     close to some neighbor anatomical structure.
> 
>     If you want to be pragmatic you could also consider the use of
>     "stoppers" as they are implemented in the SNAP application in
>     InsightApplications.
> 
> 
> 
>     Regards,
> 
> 
>     Luis
> 
> 
> 
>     ---------------------
>     conn sullivan wrote:
> 
>      > Hello Luis,
>      >
>      > What are the ITK filters other than Sigmoid that can be used to
>     generate
>      > speed images?
>      > In some cases, gradient magnitude + sidmoid filter is! not
>     helping me due
>      > to gaps produced on boundaries of gradient magnitude. This leads to
>      > leackage in the final output.
>      >
>      > Thanks,
>      > Conn
>      >
>      >
>      >
>      > */Luis Ibanez /* wrote:
>      >
>      >
>      >
>      > Hi Conn,
>      >
>      >
>      > The speed image computation starts in :
>      >
>      > Insight/Code/Algorithms/
>      > itkSegmentationLevelSetimageFilter.txx
>      >
>      > The computation is delegated to the SegmentationFunction
>      > by invoking its method:
>      >
>      > CalculateSpeedImage();
>      >
>      > This method is implemented in the file
>      >
>      > Insight/Code/Algorithms/
>      > itkCannySegmentationLevelSetFunction.txx
>      >
>      > in lines 30-38.
>      >
>      >
>      > The speed image is computed as a distance map to the Canny
>      > edges in the method CalculateDistanceimage() in lines 81-102
>      > of the same file.
>      >
>      >
>      >
>     ! >
>      > Regards,
>      >
>      >
>      >
>      > Luis
>      >
>      >
>      >
>      >
>     -----------------------------------------------------------------------------
>      > conn sullivan wrote:
>      >
>      > > Hello Luiz,
>      > >
>      > > I appreciate your reply. I get the speed image now. I have one more
>      > > question. Are speed values taken directly ! from these speed
>      > image. Also,
>      > > what is z(x) computed for canny edge filter. What part of code from
>      > > .cxx file performs this calculation?
>      > >
>      > > Thanks,
>      > > Conn.
>      > >
>      > > */Luis Ibanez /* wrote:
>      > >
>      > >
>      > > Hi Conn,
>      > >
>      > >
>      > > Thanks for posting the output of the two readers.
>      > >
>      > > We tracked this problem down and found that what you are missing is
>      > > to invoke the "GenerateSpeedImage()" method in the Canny level set
>      > > filter ! before you invoke GetSpeedImage().
>      > >
>      > > Note that the GenerateSpeedImage() method must be called after
>      > > runninig the filter with Update().
>      > >
>      > >
>      > > For your convinience we added this code to the example in
>      > >
>      > > Insight/Examples/Segmentation/
>      > > CannySegmentationLevelSetImageFilter.cxx
>      > >
>      > >
>      > >
>      > > In order to get the new version simply update your CVS checkout
>      > of ITK,
>      > ! > or go to the CVS-Web portal and download the new version of this
>      > file:
>      > >
>      > >
>      >
>     http://www.itk.org/cgi-bin/viewcvs.cgi/Examples/Segmentation/CannySegmentationLevelSetImageFilter.cxx?rev=1.27&root=Insight&view=log
>      > >
>      > >
>      > >
>      > >
>      > > Regards,
>      > >
>      > >
>      > >
>      > > Luis
>      > >
>      > >
>      >
>      >
>      >
>      > __________________________________________________
>      > Do You Yahoo!?
>      > Tired of spam? Yahoo! Mail has the best spam protection around
>      > http://mail.yahoo.com
>      >
> 
> 
> 
> __________________________________________________
> Do You Yahoo!?
> Tired of spam? Yahoo! Mail has the best spam protection around
> http://mail.yahoo.com
>