Biomedical Engineering Reference
In-Depth Information
a cyst and bone CT. No clinical validation was performed. Caselles with Malladi
et al.
in [18] compared 2D geometric and 3D geodesic deformable models with
applications on 3D CT of human thighs and 3D cardiac MRI data sets. Measure-
ment of soft tissue and ventricular cavity volumes are reported but no clinical
validation was performed. A recent review of the use of geodesic deformable
models for medical image analysis is provided in [19] with comparison of per-
formance between geometric and parametric deformable models.
2.2.5
Tuning Level Set Speed Functions
for Segmentation
The main problem of the boundary-based level set segmentation methods is re-
lated to contour leakage at locations of weak or missing boundary data informa-
tion. An illustration of the phenomenon is provided in Fig. 2.4 for segmentation
of a high-resolution abdominal MRI slice.
Several efforts have been performed to add stopping criteria on the entire
front [20, 21] and local pixel freezing rules [21], or combine gradient with region
Figure 2.4:
Leakage of level set deformable model at location of weak edges
with gradient-based speed terms. (a) T2-weighted abdominal MRI with region
of interest selected to contain subcutaneous fat. (b) Edge map derived from
gradient computation to define speed term. (c) Leakage of front outside the fat
compartment at two locations due to interstices with poor edge contrast. The
level zero curve used to initialize the segmentation is displayed with a thick
line.
