Information Technology Reference
In-Depth Information
Fig. 7 The separation of each VB in a data set. Two choices are given to the user: The Manual
and automatic options. Each option has its own advantages and disadvantages which are described
the section. a An image which has 3 adjacent VBs. b The manual separation process with 6 points
selected by a user. c The automatic separation method which was proposed by Aslan et al. in [
16
]
The advantage of automatic separation is to eliminate user interaction(s).
However, there are two disadvantages: (i) increased error, (ii) current methods in
the literature have higher execution time respect to the manual methods. To give the
user his own choice, two methods are described.
2.3.1 Manual
After the spinal cord, processes, and ribs are extracted roughly, we need to separate
adjacent VBs in order to embed the shape model to the image domain. In the
manual separation, simple manual annotations are needed to specify the cut-points
of VBs. For instance, if there are three VBs in the dataset, six points are annotated
on the image. In the experiments, the average execution time to separate 12 adjacent
VBs is 18 s. This timing may still not be optimum one, however, with manual
annotations there should not be any possible data loss. In the next section, the
automated separation process, which Aslan et al. previously published in [
16
], is
described. It should be noted that segmentation accuracy is measured when the VBs
are separated manually.
2.3.2 Automatic
In this section, a 3D framework to separate vertebral bones in CT images without
any user intervention [
16
] is used. To separate the VBs, the previously developed
approach based on 4 points automatically placed on cortical shell is used. An
example of separation and segmentation of a VB is shown in Fig.
8
c. After the
spinal cord is extracted; the approximate centerline of VB column is obtained.
These seeds are placed using the relatively higher gray level intensity values of the
cortex region.
