[Insight-users] Generalizing vesselness with respect to dimensionality and shape

[Oleksandr Dzyubak]

Dear ITKers,

Recently I decided to give a try to "Generalizing vesselness with 
respect to dimensionality and shape" from IJ
http://www.insight-journal.org/browse/publication/175
with respect to various object recognition at different scales.
To make life easier, I generated an image with the white noise and six 
Gaussian  3D lines
with diameters 20, 10, and 5 (isotropic voxels with1 mm size).
So everything is "homogeneous, isotropic, and symmetric".
The package comes with a set of tools and I chose "Frangi approach" at 
this moment.
After playing with steer parameters, I was able to extract all six lines.
However the extracted objects had shapes that were not "perfectly" 
rounded as expected but
rather having "3/4 sun eclipse" like shapes. Well, at first I though 
that it is the noise what causes
such a behavior. So I regenerated the same images without noise. 
But the result was the same "3/4 sun eclipse" like shapes.
 From the article by Frangi et al, it seems that shapes, positions,
and diameters have to be preserved if scale was chosen correctly (to 
some extend, of course).

Am I missing something? What I am doing wrong?

Thanks for your help.

Alex