Biomedical Engineering Reference
In-Depth Information
community is to find suitable algorithms to accurately and reproducibly segment
anatomical structures for clinical studies.
Traditional low-level image processing techniques perform operations using
local image statistics, producing localized patterns that need unification to form a
meaningful segmentation. However, in most cases it leads to incorrect connected
boundaries due to a lack of sufficient statistics in most regions. Moreover, the
above-mentioned artifacts immensely bias the local statistics, making it impossible
to generate anatomically correct structures. As a result, these techniques require
a considerable amount of manual intervention to generate a meaningful structure,
making it a tedious process, and one prone to operator subjectivity.
On the other hand, the use of global properties like intensity values or compact
geometric models is also not always possible since these properties themselves do
not necessarily have a one-on-one mapping with an anatomical structure or a
desired region of interest. A methodology that can encapsulate local statistics in a
global framework might prove to be a better alternative in this respect. Deformable
models [7-11] comprise a step in that direction. The main idea of these models is
that of using local statistics to deform a global geometric model. Through the last
two decades, deformable models have been a promising and vigorously researched
approach to computer-assistedmedical image analysis. The source of the immense
potential for the use of deformable models in segmentation, matching, and tracking
of anatomic structures in medical images lies in its bottom-up approach, which
exploits features derived from local image statistics along with a priori knowledge
about the location, size, and shape of these structures [12]. This allows a high range
of variability of these models to accommodate significant variation in biological
structures.
The active contour model, commonly known as the snakes model, proposed
by Kass et al. [10], defines a parametric framework for a curve that deforms under
the action of local image statistics to conform into the perceived boundary of
the structure in an image. For the last two decades, the active contour model
has found widespread application in many fields of medical image segmentation
and has undergone immense development in terms of its theoretical insight, as
well as making itself more flexible and adaptable. This chapter tries to capture the
evolution of this model and its use inmedical image segmentation. Organization of
the chapter is as follows: Section 2 provides the basic theory of an active contour
and explains the underlying physics. The confluence of geometry and image
properties is also explained in this section and the effects of each of the properties
are explored. Section 3 describes the evolution of the snake model to address
the requirements of medical image analysis applications. Section 4 describes the
inclusion of a-priori information within the snake framework. Section 5 deals with
the topological adaptability of the snake. We conclude with some discussion in
Section 6.
