[Insight-users] Re: Cellular images : Level Set segmentation

[Luis Ibanez]

Hi XueQuan

1) If you look at the signature of the GetVoronoiMap() method,
    you will notice that the image type of the Voronoi map is
    defined by the second template parameter of the filter.

    Therefore, if you anticipate that there will be more than 255
    regions in the Voronoi partition, then you should use an image
    type with larger pixel capacity. For example, an image of pixel
    type "unsigned short", or even larger "unsigned long".


2) Note that you use the seed in a preprocessing filter
    (the FastMarching filter), you should still put the seeds inside
    the [biological] cells, grow from them initial level set, that
    outgrow the cells (e.g. that are larger than the cells), then
    you negate this initial level set and pass it as input to the
    ShapeDetection or the GeodesicActiveContours level set, so it
    will now progress by shrinking towards the [biological] cells.



  Regards,


      Luis


-----------------------------
qxue at jouy.inra.fr wrote:
> Hi, Luis:
>     Thanks for your help, and now I have finished part of the program based on
> ITK.
>     May I have a discussion about my question further?
> 
> 
>>1) You can take the Voronoi map image produced by
>>    the Danielsson or the Maurer distance map filters.
>>
>>2) For every marker, theshold the voronoi map, so that
>>    you end up with a mask that has the voronoi region
>>    corresponding to a single nucleus.
>>
> 
>       My image is about cell segmentation, so sometimes there are more than 255
> cells in a picture. For this condition, will some of the Voronoi maps use the
> same grey level to distinguish each region?
> 
> 
>>3) Use this binary region to mask the speed image that you
>>    compute from the gradient magnitude of the original image.
>>    Pass the resulting masked image as the feature image to
>>    the level set. In this way you are restricting the level
>>    set to grow into the voronoi region at most.
>>
>>
>>
>>About your other question,
>>
>>If you negate the initial level set, then you can have
>>it behaving as a shrinking front instead of an expanding
>>front.
>>
> 
>   Until now, I use the one seed or more which is located in the center of the
> objects level set method to get the segmentation. if I negate the initial level
> set, should I change the seed? If yes, what should I do? I want to find whether
> ITK supports the circle or rectangle shrinking similar to the referece paper I
> sent you several days ago. Should I select many seeds in a circle shape and
> then let level set shrink to the center?
> 
>        best regards,
> XueQuan
>