[Insight-users] Fuzzy parameters.

[Luis Ibanez]

Hi Rodrigo,

Thanks for pointing this out.

1) A bug was found on the RegionGrowingSegmentation.
    The SimpleFuzzyConnectedness filter was not
    receiving the coordinates of the seed point.
    This has been fixed today in CVS.  Please update
    your cvs checkout in order to get this modifications.


About the parameter setting. Here are two examples,
one in 2D and another in 3D. Hopefully by playing
with them you will get a feeling on the effect of
the parameters

2) Example in 2D

    - Run RegionGrowingSegmentation2D
    - Load the image BrainProtonDensitySlice.png
    - Smooth with CurvatureFlow only for one iteration
    - In the homogeneous image, click on one of the
      ventricles. (the white structures close to the
      center of the brain).
    - The GUI reports the pixel value at the seed point,
      use this value as estimation for the Mean in
      the FuzzyConnectedness filter. (it should be around
      "230")
    - Use "30" as estimation of the Variance. Note that
      this is Variance, not standard deviation... so you
      should think in squared units.
    - Set the threshold to 0.5
      Note that the GUI changed today, so the Threshold
      value is now exposed.

    Click on the display button, you should see the ventricle
    with a number of holes on it.  Play first reducing the
    threshold value, that should tend to fill in the holes.
    Play then with the variance value. You should be able to
    go as high as 500 in the variance and still get reasonable
    segmentations.  The value of the mean is more critical
    than the value of the variance.

    In order to get a feeling for mean estimations you could
    make many clicks in the region you want to segment, and
    from the pixel values manually estimate the mean of the
    region.


3) Example in 3D

    - Run RegionGrowingSegmentation
    - Load the BrainWeb image

            brainweb1e1010f20.mha

      these images are available from our FTP site.
      Look at the "Data" link in the ITK home page for
      instructions on how to get this data.

    - Run only one iteration of Curvature Flow on it.

    - Display the Homogeneous image, and move to slice 94
    - Again click on one of the ventricles. For example in
      coordinates (98,112,94).

    - Set the mean estimation to 230
    - Set the variance estimation to 30
    - Set the threshold to 0.5

    - Click on the Display button in front of FuzzyConnectedness.
    - The segmented volume should appear, move to slice 94 in order
      to see the same structure as where you selected the seed point.
      Since the structure will have hole, you may want to lower the
      affinity threshold to 0.15. This will produce a more compact
      segmentation.



4) Note also that there is an example on the use of the
    SimpleFuzzyConnectednessImageFilter, under:

       Insight/Examples/Segmentation/
                   FuzzyConnectednessImageFilter.cxx

    In this example, the mean and variance of the object
    to be segmented are estimated by using the Confidence
    Connectedness image filter.  This filter does a rough
    first segmentation of the object, and has the advantage
    of returning the values of mean and variance corresponding
    to pixels in the segmented region.



Please let us know if you have further questions.



   Regards,



      Luis




-------------------------------
Rodrigo Trujillo wrote:

> Hi,
>  
> I am trying the application RegionGrowingSegmentation.exe of 
> InsightApplications. I choose Fuzzy
> Connected, I got only an empty image.
>  
> I read itkSoftwareGuide and the 
> SimpleFuzzyConnectednessScalarImageFilter doxygen
> documentation. But nothing explains the roles of fuzzy's main 
> parameters: mean, variance and
> threshold.
>  
> I choosed Connected Threshold for segmentation to see if there was an 
> unexpected problem
> throghout the previus part, but it worked fine.
>  
> I can't use Fuzzy without a clue of what their parameters are. It's 
> impossible to learn only
> by trying aleatorios combinations.
>  
> Should one explain the roles of these parameters?
>  
> Thanks.
>  
> Rodrigo Trujillo
>