Medical Image Processing - Signal Processing Systems - page 357

application or extension to three dimensions can be extraordinarily time-consuming.
Novel architectures for image preprocessing are an area of active research and
development. Of special note is the field programmable gate array (FPGA)-based
generalized architecture by Dandekar et al. [ 5 ] with an efficient voxel access scheme
customized for neighborhood operations-based image preprocessing such as 3D
median filtering and 3D anisotropic diffusion filtering.

6 Image Segmentation

Image segmentation is a class of computer vision algorithms that involves iden-
tifying and subdividing an image into its components or constituent parts called
segments. Image segmentation can be thought of as assigning a label to every pixel
or voxel in an image, such that pixels/voxels with the same label share a certain
specific characteristic or computed property such as color, intensity, or texture. Pix-
els/voxels having identical labels are grouped into similar segments, which typically
are the objects and boundaries (lines, curves, etc.) within images. Such identification
of the image segments often is the first important step in performing more complex
image analysis operations. Separating the image segments also aids in effective
visualization of complex image data by simplifying the image information into sets
of spatially, structurally, and/or functionally correlated primitives.
In medical imaging applications, image segmentation typically refers to the iden-
tification of known anatomic structures from images. Structures of interest include
organs or parts of organs, vascular structures, skeletal components, or abnormalities
such as tumors and cysts. Segmentation is a key step to enable detection/diagnosis
(by locating tumors and other pathological conditions, measuring volumes of tissues
and cavities, etc.), image-guided interventional procedures (by creating accurate
2D/3D anatomical maps specific for individual patients), and other applications such
as staging and monitoring treatment responses.

6.1 Classification of Segmentation Techniques

A wide range of segmentation methods continues to be developed for addressing
specific problems in medical applications. In contrast to generic segmentation
methods, methods used for medical image segmentation are often application-
specific; as such, they can make use of prior knowledge for the particular objects
of interest and other expected or possible structures in the image.
A commonly used approach to classify segmentation methods is based on the
image primitive being identified by a given method. Basic image primitives in
this sense are pixels/voxels, regions/volumes, and edges/surfaces. Accordingly,
segmentation techniques can be classified as