Biomedical Engineering Reference
In-Depth Information
of one plane will minimize the computational cost of the procedure, ensuring that
the segmentation time is short. Since the prostate is approximately ellipsoidal,
the cross-sectional plane passing through the approximate center of the prostate
will result in a large and smooth boundary, which is better than the cross-sectional
planes near the ends of the prostate.
4.2.4. Propagation in two opposite directions
The modified slice-based 3D prostate segmentation method described above
can generate a smooth prostate surface that matches the actual prostate boundary in
most cases, i.e., when the segmentation errors occur near the end of the propagation.
However, it is possible that segmentation errors will occur near the beginning of
the propagation, usually due to the loss of prostate boundary contrast produced
by shadowing from calcifications, brachytherapy seeds, or needles. To overcome
this problem, the modified slice-based 3D segmentation method is repeated in
the opposite propagation direction. Thus, the slices near the beginning of one
propagation (e.g., the clockwise direction with respect to the coronal view) will
appear near the end of the other propagation in the anti-clockwise direction.
It is important to note that the worst case will occur when segmentation errors
occur at both ends of the propagations. To handle this case, the bidirectional
contours can be combined, as described by Ding et al. [44], to obtain the final
optimal contour.
4.2.5. Evaluation
Our slice-based 3D segmentation method using a continuity constraint im-
plemented with an AR model has been used to segment 3D TRUS images of 15
patients' prostates obtained using the 3D TRUS system described above. Some
of the 3D TRUS images are shown in Figure 15, and others can be found in [44].
Each 3D TRUS prostate image was segmented five times manually and using four
slice-based segmentation algorithm methods: the original as described by Wang
et al. [41] (SB), clockwise (CP), anti-clockwise (AP), and a combination of both
CP and AP (OPT).
Figure 16 shows a segmentation example of the 3D TRUS prostate image
shown in Figure 14b. Figure 16a shows the segmentation results in the 0th slice
(i.e., the initial slice) and the three contours obtained using the CP (green), AP
(blue), and OPT (purple) methods. It is evident that these contours are almost the
same as the contour produced by manual segmentation red. Figure 16b shows the
segmentation results in the 17th slice. In this slice, the OPT contour (purple) on
the left side of the prostate is based on a selection rule described in [44] of the CP
and AP contours, and the OPT contour on the right side was obtained from the AP
contour alone. Figure 16c is the segmentation result in the 59th slice. The OPT
contour on the right side of the prostate is based on a selection rule described in
