Biomedical Engineering Reference
In-Depth Information
(a)
(b)
Figure 12.
Images showing the performance of the inertial snake on (a) computer-generated
phantom image and (b) ultrasound image of an artery of the lower limb [26]. See attached
CD for color version.
fit into the process. As the contour itself is evolving under the effect of geometric
physical constraints, the framework in its very definition imposes a geometric
smoothness into it. Although this is effective in many situations, many other cases
require relaxation of the physical constraints for appropriate segmentation. While
in tracking geometric deformation in smooth structures the curvature is helpful,
in segmentation of structures like white matter or convex structures the effect of
curvature needs to be reduced. A study on the discrete computation and effect of
curvature is presented in [15].
A potential way of controlling snake movement to conform into image bound-
aries and to avoid leaking is by incorporating an adaptable propagation force that
modifies itself as it moves from a homogeneous to a nonhomogeneous region [26].
At this point, this is done by using an inertial snake, where the propagation term
of active contour deformation is controlled by its distance from the initial contour.
In that approach, the balloon force modifies itself and slows down at a rate directly
proportional to the deformation rate between the successive iterations. Thus, in
homogeneous region the force is higher, supporting faster propagation. But as the
contour approaches the high-gradient region, the force reduces itself, thus discour-
aging propagation. The uncontrolled motion of the balloon force is thus controlled
by the homogeneity and gradient information. Figure 12 illustrates the result of
the inertial snake on computer-generated phantoms and ultrasound images.
Other significant attempts to improve the snake model using gradient-based
information and internal energy has been pursued by other researchers [27, 28].
However, the main limitation remained its dependence on local image statistics,
which is not reliable in many situations. Mostly for medical image analysis and
