[Insight-users] vessel segmentation

[Bryn Lloyd]

Hi,


I am trying to segment small vessels from MRA images (in mice).

The images typically have a spacing of [0.1, 0.1, 0.5] mm.

I find I can detect the vessels which are oriented more or less parallel 
to the z-direction quite well. The in-slice (x-y) vessels are much more 
difficult.

One of the problems is that a vessel might e.g. be at position [10,10] 
in one slice and at [15,15] in the next. This large "jump" is due to the 
large spacing in z-direction.


Can anybody suggest an approach which can handle such situations?

I have tried region growing on "vesselness" feature images, and thought 
about using the CURVES algorithm. The problem with CURVES (I expect) is 
that there will be holes where the level set can leak out.

In addition, the size of the vessels can be very small (approximately 
2-5 pixels diameter). According to a mail from Luis, CURVES needs at 
least 5 pixels to work well. Can one trick CURVES to work also for 
smaller vessels, e.g. through (in-slice) re-sampling using nearest 
neighbor interpolation?


Thanks
Bryn