[Insight-users] How dose ITK leve set work on physical space

[Luis Ibanez]

Hi Baoyun,


The effect of enabling the use of image spacing in the Level Set 
framework occurs at the level of the itkFiniteDifferenceImageFilter.

If you look at the code in


      Insight/Code/Common/itkFiniteDifferenceImageFilter.txx

lines 44-57.


You will find that the effect of the spacing is not really to make
bigger (nor smaller) steps when propagating the contour of the level
set, but to compensate for the (potential) anisotropy of the image.

In practice, when propagating the zero set, it will be restricted to
not moving more than one pixel away from its current position at
every iteration.

To put it short:

      The image spacing only matters for anisotropic images.



    Regards,



        Luis


-----------------
Baoyun Li wrote:
> Dear Lui and all:
> 
> I am confused by muti-scale frame work with physical space processing in 
> ITK. Let take Level set for segmenation as example, the original data is 
> 1mm, and I will do the 3 reolution level for level set (4,2,1 mm) 
>  
> 1 I fully understand what happen in voxle space. we have intial level 
> set, at high resolution, the distance of initial levels set may be too 
> far to the true contour( let say 10mm or 10 voxle), then we go to 4mm, 
> and the distance is only 2.5 voxels. So muti-scale processing help the 
> level set to converge to the truth quickly.
>  
> 2. But if we process it in the ITK framework or say physical space, even 
> at low resolution, the distance is still 10mm and is not reduced. How 
> this muti -scale scheme helps to speed up the algorithm?
>  
> 3 I guess, even we say ITK is in physical space, but in real processing, 
> e.g, (at 4mm level), the program will only work on the grid point. But 
> the programm knows clearly that is 4mm voxle distance. The step is 4 
> time bigger than high resolution case.( I guess).
>  
>  
> I understand this question may be too naive, but I am really confused.
>  
>  
> Thanks
>  
> Baoyun
>