Biomedical Engineering Reference
In-Depth Information
describe their segmenting curve. Leventon et al. [56] proposed a less restrictive
model-based segmenter. They incorporated shape information as a prior model
to restrict the flow of the geodesic active contour. Their prior parametric shape
model was derived by performing PCA on a collection of signed distance maps of
the training shape. The segmenting curve then evolves according to the gradient
force of the image and the force exerted by the estimated shape. In 2002, Mitchell
[61] developed a model-based method for three-dimensional image seg-
mentation. Comprehensive design of a three-dimensional (3D) active appearance
model (AAM) was reported for the first time as an involved extension of the AAM
framework introduced by Cootes et al. [62]. The model's behavior is learned from
manually traced segmentation examples during an automated training stage. In-
formation about shape and image appearance of the cardiac structures is contained
in a single model. The clinical potential of the 3D AAM is demonstrated in short-
axis cardiac MR images and four-chamber echocardiographic sequences. The
AAM method showed good agreement with an independent standard using quan-
titative indices of border positioning errors, endocardial and epicardial volumes,
and left-ventricular mass. The AAM method shows high promise for successful
application to MR and echocardiographic image analysis in a clinical setting. The
reported method combined the appearance feature with the shape knowledge, and
it provided robust matching criteria for segmentation. However, this method needs
to set up point correspondence, and it makes the procedure complicated.
The authors of [37] adopted implicit representation of the segmenting curve
proposed in [36] and calculated the parameters of the implicit model to mini-
mize the region-based energy based on a Mumford-Shah functional for image
segmentation [58]. The proposed method gives a new and efficient framework for
segmenting an image contaminated with heavy noise and delineating structures
complicated by missing or diffuse boundaries. In the next section we will give a
detailed description of this framework in a 2D space. The flowchart of the method
is depicted in Figure 16.
et al.
5.1. Shape Model Training
, each of which with 1 as
object and 0 as background. In order to extract the accurate shape information,
alignment has to be applied. Alignment is a task to calculate the following pose
parameters
p
=[
abhθ
]
Given is a set of binary images
{
B
1
,B
2
,...,B
n
}
T
, and correspondingly, these four parameters are for
translation in
x
,
y
, scale, and rotation:
10
a
01
b
001
h
00
0
h
0
00
h
cos(
θ
)
−
sin(
θ
)0
.
T
(
p
)=
sin(
θ
)
cos(
θ
)0
(30)
0
0
1
The strategy for computing the pose parameters for
n
binary images is to use
the gradient descent method to minimize the special designed energy functional
