[Insight-users] Problems on using the GeodesicActiveContourSegmentation Application

[Luis Ibanez]

Hi Kai,

Thanks for letting us know that the
parameters worked for you too.

Isn't it great when things are reproducible ?  :-)

About the scaling parameters, your observation
is correct.  These three parameters get normalized
inside the filter, so what actually count are their
relative ratios.

A typical strategy is to start with low values of
propagation and advection and only increase their
weight if you observe that the front is not moving
fast enough.  You don't want to start with large
values of the scaling for propagation and advection
since they tend to produce leaks.


   Regards,


     Luis



----------------
Kai Li wrote:

> Hi Luis,
>    Thank you very much for your clear answer. I followed the parameters
> you gave and it worked well. Except for not properly setting the scaling
> parameters, I think the problem I had was also due to that the RMS error
> was not little enough either.
> 
>    About the scaling parameters, I tried the combination of 1.0, 20.0,
> 20.0 and 0.1, 2.0, 2.0, it seems that the result is basically the same.
> So is it the case that the scaling parameters are proprotinal
> equivalement? Are there some generic rules to set these parameters?
> 
> Thanks,
> Kai
> 
> On Wed, 7 Apr 2004, Luis Ibanez wrote:
> 
> 
>>Hi Kai,
>>
>>Here are the parameters that you can use
>>with the GeodesicActiveContourSegmentation
>>application (in InsightApplications), in
>>order to segment the ventricles from one
>>of the BrainWeb datasets.
>>
>>
>>1) Load the BrainWeb image:
>>
>>         brainweb1e1a10f20.mha
>>
>>2) In FastMarching
>>    2.1) Set the distance to 3.0
>>    2.2) Set Stopping value = 10.0
>>
>>4) In Gradient magnitude keep sigma = 1.2
>>
>>5) In Sigmoid keep alpha = -1, beta= 5.0
>>
>>6) In GeodesicActive contour do:
>>
>>    6.1) RMS error  = 0.01
>>    6.2) Iterations = 200
>>    6.3) Curvature scaling   =  1.0
>>    6.4) Propagation scaling = 20.0
>>    6.5) Advection scaling   = 20.0
>>
>>7) Place seed point at locations
>>    (or close to them)
>>
>>    7.1)   ( 67,  87,  87)
>>    7.2)   (113,  87,  87)
>>    7.3)   ( 85, 135,  87)
>>    7.4)   ( 95, 135,  87)
>>    7.5)   ( 78, 108,  97)
>>    7.6)   (101, 108,  97)
>>
>>
>>The reason why you may be getting always
>>a sphere is that the weight of propagation
>>and advection terms are too low in your
>>case, so the front is not moving enough.
>>
>>The VTK visualization button is something
>>that was never finished in this application.
>>So it is normal (although it is not good)
>>that the displayed window is all white.
>>
>>If you really want to see a volume rendering
>>of the resulting segmentation, is is much
>>easier now to use VolView and  its ITK Geodesic
>>Active Contour plugin.  You can download the
>>free version of VolView from:
>>
>>http://www.kitware.com/products/volview.html
>>
>>--
>>
>>------------------------------------
>>IMAGINE: If one day published papers
>>will have all the source code, data
>>and the parameters needed in order
>>to reproduce results...
>>------------------------------------
>>
>>
>>
>>   Luis
>>
>>
>>
>>
>>--------------
>>Kai Li wrote:
>>
>>
>>>Hi,
>>>  I tried the GeodesicActiveContourSegmentation application for a brainweb
>>>dataset. It seems to work fine with segmentation of white matter. But I
>>>haven't made it work with ventricles. What I did was just putting a seed
>>>point within a ventricle and leave all parameters default except for the
>>>distance of FastMarching filter,  the RMS error and iterations of the
>>>GeodesicActiveContour filter. I set the distance of the FastMarching
>>>filter to be just "1.0". The problem is that no matter how I decrease the
>>>RMS error and increase the iterations, the result is always a small sphere
>>>like shape around the seed point.
>>>
>>> Another problem is about the VTK Render. After the Threshhold filter is
>>>executed, the color in the "VTK Render" button is still red. If I click
>>>on the button, I just get a blank window. The fltkImageViewer(the
>>>"Display" button) works fine though. I think I have succesfully installed
>>>the VTK library.
>>>
>>> Any help is highly appreciated.
>>>
>>>Kai
>>>_______________________________________________
>>>Insight-users mailing list
>>>Insight-users at itk.org
>>>http://www.itk.org/mailman/listinfo/insight-users
>>>
>>
>>
>>
>>_______________________________________________
>>Insight-users mailing list
>>Insight-users at itk.org
>>http://www.itk.org/mailman/listinfo/insight-users
>>
> 
>