Biomedical Engineering Reference
In-Depth Information
For the optimization example discussed above, each of the 15 images was
segmented 100 times for a particular value of
σ
. This would require anywhere from
3.3 to 10 hours if done manually by a user. The time requirement to segment many
images many times is prohibitive if virtual operators are not available. Although
the construction of each virtual operator requires some investment of time, the
virtual operator can then be applied as many times as needed without the need for
the user to be present.
4.
3D SEGMENTATION
In many therapy and surgery applications such as prostate brachytherapy and
cryosurgery, the 2D segmentation techniques described in Section 2 are inade-
quate, as they do not provide the required 3D description of the prostate. To obtain
the 3D information, such as the volume or the 3D boundary of the prostate, a 3D
prostate segmentation method must be used. 3D prostate segmentation methods
can be categorized into two classes: propagation of a 2D slice based-segmentation
to 3D, and direct 3D segmentation. In Section 3.1, a 2D slice-based 3D prostate
segmentation method is described [41, 42]. In order to overcome the potential seg-
mentation error accumulation problem observed in the 2D slice-based 3D segmen-
tation method, a modified slice-based 3D segmentation method using continuity
constraint implemented with an AR model is described in Section 3.2. In Section
3.3, a direct 3D prostate segmentationmethod using an ellipsoid deformable model
is described [43].
4.1. Slice-Based 3D Prostate Segmentation
The basic idea of 2D slice-based 3D prostate segmentation method is to solve
the 3D prostate segmentation problem by re-slicing the 3D prostate image into
a series of uniformly spaced contiguous 2D images and segment the prostate in
these 2D images. Differentiating from the 2D prostate segmentation method that
requires manual initiating the initial contour of the prostate in each 2D image, the
slice-based 3D segmentation method only requires initialization of one contour
in an initial slice, then, iteratively propagating the refined contour to its adjacent
slice and deforming it to the boundary of the prostate. The procedure is repeated
until all slices are segmented. After the segmentation of all re-sliced 2D images
is obtained, a 3D surface mesh of the prostate is reconstructed from the contours
in all 2D slices. The details of the slice-based 3D prostate segmentation method
is described in the following steps:
Step 1: Re-slicing.
First we re-slice the 3D US image of the prostate into a series
of slices using one of two possible methods [41]. In the first — called
parallel re-slicing — the prostate is cut into a series of parallel slices with
a uniform spatial interval, usually in the transverse plane (Figure 11a).
