Information Technology Reference
In-Depth Information
using the level sets. They minimized a certain function to estimate the transfor-
mation parameters.
In this chapter, we present universal and probabilistic shape based segmentation
methods that are less variant to the initialization. Contribution of this chapter can be
generalized as follows:
This chapter solves problems caused by noise, occlusion, and missing infor-
mation of the object by integrating the prior shape information.
In this chapter, the conventional shape based segmentation results are enhanced
by proposing a new probabilistic shape models and a new energy functional to
be minimized. The shape variations are modelled using a probabilistic functions.
The proposed shape based segmentation method is less variant
to the
initialization.
To optimize the energy functional, the original ICM method, which was orig-
inally proposed by Besag [
13
], is extended by integrating the shape prior. With
integrating the shape model to the original ICM method, possible local mini-
mums of the energy functional are eliminated as much as possible, and enhance
the results.
One of the most important contributions of this study is to offer a segmentation
framework which can be suitable to the clinical works with acceptable results. If
the proposed method in this study is compared most published bone segmen-
tation methods, the large execution time is reduced effectively.
Many works are restricted to the speci
c regions of spine bone column as such
lumbar, thoracic, and others. In this study, there is no any region restriction, and
the proposed framework is processed on different regions.
The proposed framework and the new probabilistic shape model extract the
spinal processes and ribs which should not be included in the bone mineral
density measurements.
This work is not dependent on any identi
cation step thanks to the new uni-
versal shape model and its embedding step.
Next section details the proposed methods.
2 Proposed Framework
In this section, we describe the proposed methods. First, the general theoretical idea
of the proposed frameworks is described. Then, two pre-processing steps, the spinal
cord extraction and VB separation, are described. We present three methods which
differ mostly in the shape modeling and optimization parts.
