Biomedical Engineering Reference
In-Depth Information
difference between the zero level and the statistical prior contours, and embedded it
in the evolution process. Chen et al. [106] made it applicable to a linear transform.
They all got a more robust result than with the traditional method. However,
calculating the difference between the zero level and prior statistical contour at
each evolution step is time consuming, and a scalar of the similar contour metric
cannot guide the evolution directly. In the following, we present a new algorithm
that represents the region and shape prior knowledge in a form of a speed field.
The speed vector can drive the zero level set directly to the ideal cardiac valve
contour.
6.3. Geodesic Snake Guided by Prior Knowledge
Given a region or shape priors, the information can be folded into the seg-
mentation process. We here represent the region and shape prior as speed fields
and embed them to the level set evolution equation to pull the surface to the ideal
contour, which could reduce the manual parts procedure to a great extent.
Our approach is based on a geodesic snake [49]:
∂φ
∂t = u ( x )( k + v 0 ) |∇
φ
| +
u
·∇
φ,
(34)
where u ( x )= −|∇
, and v 0 is an image-dependent balloon force added
to force the contour to flow outward. In this level set framework, the surface φ
evolves at every point perpendicular to the level sets as a function of the curvature
at that point and the image gradient.
We add new speed items to the evolution equation of the geodesic snake. The
new prior knowledge item forms a total force with an internal and external force
of the original image and drives the zero level set to converge to the ideal contour.
There are two levels of the additional prior force: one is the lower level of the
region prior constraint, which makes the zero level evolve in a certain region, and
the other is the higher level of the shape prior constraint, which makes the final
contour converge to the prior shape of object. We then obtain a new equation:
G σ
I
|
∂t = u ( x )( k + v 0 ) |∇ φ | + u ·∇ φ + F i ·∇ φ,
∂φ
(35)
where F i is the region, shape, or other prior knowledge forces. The speed force
is more direct and efficient than the similarity metric of the contour. However, it
is difficult to transform the prior knowledge into a speed field. The algorithm for
this is presented in detail as follows.
6.3.1. Region prior-based geodesic snake
The movement of a cardiac valve is very complex. It moves with the beating
of the heart, turns around the valve root, and creates a great deal of distortion by
Search WWH ::




Custom Search