jorgen.kongsro@[EMAIL PROTECTED]
wrote:
> I have a stack of CT images from a whole body scan of live pigs. I
> want to remove internal organs and perform a virtual dissection of
> each pig. The image stack is a data matrix or cube of size 512 x 512 x
> 1200, where each image is a 512 x 512 matrix, and the total number of
> images per pig is approx. 1200.
>
> I have tried som different strategies so far:
>
> 1. Thresholding by gray values
> Each tissue have different gray values, and it is easy to separate
> soft tissues (fat, muscle, internal organs) from cartilage (bone). The
> problem is to separate the soft tissues from each other, especially
> muscle tissue from internal organs like kidney, liver, stomach etc.
> They have the same gray values.
>
> 2. *****on / dilation
> The next step was to segment the tissues based on *****on / dilation.
> This works well when there are solid or clear boundaries between
> internal organs and muslce tissue separated by the body cavity. But it
> is not possible to separate when the boundaries are more diffuse, i.e.
> kidneys and liver connected to the body cavity.
>
> 3. Active contour / "s****s"
> Recently, I have tried to use active contours or "s****s" by inserting
> seed points along the body cavity. This have worked pretty well, and I
> am able to perform the segmentation in a semi-automatic way. For 3D
> segmentation along the saggital or coronal plane of the animal, I have
> extrapolated the seed points and make new ones when the anatomy of the
> animal changes. Total process time for this procedure is approx 8-10
> minutes per pig.
>
> I want to reduce process time and automate this procedure even further
> by reducing the manual labour of seed point according to anatomy in
> the saggital / coronal direction. I wish I could attach a picture, but
> you can view some image examples at this external web site (google
> search: CT image, pig):
>
> http://cmiss.bioeng.auckland.ac.nz/development/examples/a/as/index.html
>
Hi
Interesting problem.
I have previously worked with similar problems see e.g.
http://www2.imm.dtu.dk/pubdb/views/publication_details.php?id=4859
But you might consider building Active Shape Models (ASM) or Active
Appearance Models (AAM) - try to google these.
They are based on training sets so if you have several maually segmented
pigs you may be able build the model.
Good luck,
Allan
|
