Biomedical Engineering Reference
In-Depth Information
(a)
(b)
(c)
Figure 20. Cardiac valve segmented by region prior-based geodesic snake: (a) image from
an echocardiographic sequence and the prior valve region; (b) result of preprocessing; (c)
final segmenting result with region prior-based geodesic snake.
We erode the prior region Ω and obtain Ω , and then fill the segmentation result
with foreground color outside Ω . Finally, we perform an intensity reversion and
obtain the final result.
Preprocessing is very important because noise is inevitable in an ultrasound
image. We adopted a Modified Curvature Diffusion Equation (MCDE) [107,108]
to filter the original image, which could preserve the edge and smooth the noise.
The equation is given as
u
u t = |∇
u
|∇·
c ( |∇
u
| )
.
(42)
|∇
u
|
We offer an example of a cardiac valve segmented using a region prior-based
geodesic snake in Figure 20.
6.3.2. Shape prior-based geodesic snake
Because of the intrinsic noise in an echocardiographic image and the blur
caused by movement of the cardiac valve, it is unavoidable that there will be some
segmentation errors in the final result. To get a more accurate contour, we need to
make full use of the prior shape of the heart valve in the segmenting process.
To that end, we set a speed field nearby the prior shape that directs to the
nearest point of the shape. The force F shape pulls all the points near to the prior
shape. Consider a prior contour C . We define the distance from point X to contour
C as ε :
ε ( X )= d ( X ) = min ( |
X
X I | )
X I
C.
(43)
Search WWH ::




Custom Search