Biomedical Engineering Reference
In-Depth Information
Figure 18.
Cross-sections through the image in Figure 17 along with the algorithm seg-
mentation (solid line) and manual segmentation (dotted line). (a) 3D TRUS image of the
prostate with transverse, coronal, and sagittal cutting planes indicated by (b-d), respec-
tively, to show 2D cross-sectional images. (b) Transverse cross-section of the image and
the boundaries corresponding to the plane shown in (a). (c) Coronal cross-section of the
image and the boundaries. (d) Sagital cross-section of the image and boundaries. Reprinted
with permission from the AAPM.
cutting planes to show 2D cross-sectional images. Figures 18b-d show the same
2D cross-sections, respectively, with cross-sectional contours through the manu-
ally and algorithm-segmented meshes superimposed. The manual and algorithm
contours are similar and follow the prostate boundary well in regions where the
contrast is high. In regions of low ultrasound image signal and ultrasound image
shadowing, the prostate boundary is difficult to discern in the images, and the man-
ual and algorithm-segmented contours differ from each other. This is particularly
apparent in regions near the bladder (indicated by the white arrow) and the seminal
vesicle (indicated by the black arrow), as shown in Figure 18d.
The complete direct 3D semiautomatic DDC-based segmentation procedure
required less than a minute to segment a prostate in a 3DTRUS image on a Pentium
III 400 MHz personal computer. Of this total time, approximately 5-6 sec were
required to initialize the algorithm, and 20 sec were taken up by the deformation
algorithm. Editing was required for only one 3D image, and took approximately
30 sec. In contrast, 1 to 1.5 hours were required to segment a single 3D TRUS
image manually.
