[Insight-users] Geodesic Active Contour and MRA vessel extraction

[Luis Ibanez]

Hi Fucang,

I assume you are using the GeodesicaActiveContour
application in InsightApplications... is that right ?

If this is the case,
this applications has two stages. In the first one,
it uses FastMarching for generating a LevelSet from
a set of seeds.  In a second stage, the LevelSet
produced by FastMarching is passed as initial input
to the GeodesicActiveContours filter.

You can see this tamdem as a rough and fast first
generation of a level set (FastMarching) followed by
a refinement using a higher quality but more time
consuming technique (GeodesicActiveContours).

Given that the whole process is trying to grow a
contour from a set of seed. It will be very hard
to get a segmentation of vessel using a single
seed.

Using a single seed means that you have to let the
algorithm running for a long time in order to allow
the contour to reach the other end of the vessel.
During this long time, the front will be making
pressure on the walls of the vessel that are close
to the seed position. The result is that the front
will start leaking in the places close to the seed
before other parts of it reach the ends of the vessel.

All that to say:  "You need many seeds"

Try placing as many seeds as you can all along the
medial axis of the vessel. You don't have to put
one per slice,... probably a good measure is to
put seeds spaced by a distance equal to the radius
of the vessel. Since, each seed will generate a
spherical wavefront,  in this configuration the
fronts will touch the vessel walls at the same time
they touch their neighbor seed's front.

Level set front will fuse nicely in this way.


2) Using fast marching you are generating a contour
    from seed points, so... this is already covered.


3) The level set is capable of making turns in space.
    There is no restriction on directionality at all.
    So, it should be able to walk inside any shape.

    The only restriction is that the width of the shape,
    (in this case the radius of the blood vessel)
    has to be large enough to accomodate inside twice
    the transition band of the level set.

    You can imagine this as a viscosity restriction,
    by which there is a minimum radius for a tube
    to be able to move oil inside.

3) I would suggest you to start with a syntethic
    image of the anatomical object you want to
    segment.  E.G. create a black image with a
    white cylinder inside. Select the cylinder's
    diameter to be close to the expected blood
    vessel diameter.

    Experiment first with this clean image in order
    to get a better feeling on the behavior of the
    level set. Only move to real images when you
    are comfortable with the parameter selections
    of the syntetic image.


Please let us know if you have further questions.

Fucang Jia wrote:
> Hi Luis, hi everyone,
> 
> I am interested in Geodesic Active Contour using segmentation of MRA 
> vessel. I use the data from 
> http://caddlab.rad.unc.edu/publications/Data/Branch.n000.mhd ( Thanks 
> Dr. Aylward ), the default parameters in the program 
> GeodesicActiveContour was used, but the result seems level set only 
> track a few images. I tune the parameter, but when iterator number 
> becomes big, the track depth only adds a little. So I want to know: what 
> is the art when track blood vessel from MRA using Level set ? Suppose 
> the 3d volume are smoothed fine.
> 
> Other questions I am concerned:
> 1. Should initialize one point enough? Or should use more seed points at 
> different image (in the volume) to speed up the evolution?
> 
> 2. A contour not seed point should or should not speed up the evolution 
> progress, if yes, how to feed the contour to the level set filter?
> 
> 3. Blood vessels are not always on one direction, so the blood vessel 
> can change direction at one image, like "n" character, so fast marching 
> can not used in segmentation MRA, is that right?
> 
> Thanks for any advice!
> 
> Fucang
> 
> _________________________________________________________________
> MSN 8 with e-mail virus protection service: 2 months FREE* 
> http://join.msn.com/?page=features/virus
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-users
>