Biomedical Engineering Reference
In-Depth Information
magnetic resonance (MR) imaging, or by a combination of images of the same
object acquired from different imaging modalities in general.
There are, however, many applications where there is no well-defined re-
lationship between a voxel's value(s) and the label that should be assigned to
it. This observation is fairly obvious when we are seeking to label
anatomical
structures
rather than
tissue types
. It is clear, for example, different structures
that are composed of the same tissue (e.g., different bones) cannot be distin-
guished from one another by looking at their intensity values in an image. What
distinguishes these structures instead is their location and their spatial relation-
ship to other structures. In such cases, spatial information (e.g., neighborhood
relationships) therefore needs to be taken into consideration and included in
the segmentation process.
Level set methods [37, 66, 75, 86] simultaneously segment all voxels that
belong to a given anatomical structure. Starting from a seed location, a discrete
set of labeled voxels is evolved according to image information (e.g., image
gradient) and internal constraints (e.g., smoothness of the resulting segmented
surface). Snakes or active contours [85] use an analytical description of the
segmented geometry rather than a discrete set of voxels. Again, the geometry
evolves according to the image information and inherent constraints.
In addition to geometrical constraints, one can take into account neighbor-
hood relationships between several different structures [74, 84]. A complete
description of such relationships is an
atlas
. In general, an atlas incorporates
the locations and shapes of anatomical structures, and the spatial relationships
between them. An atlas can, for example, be generated by manually segmenting
a selected image. It can also be obtained by integrating information from multi-
ple segmented images, for example, from different individuals. We shall discuss
this situation in more detail in section 11.4.3.
Given an atlas, an image can be segmented by mapping its coordinate space
to that of the atlas in an anatomically correct way, a process commonly referred
to as registration. Labeling an image by mapping it to an atlas is consequently
known as atlas-based segmentation, or registration-based segmentation. The
idea is that, given an accurate coordinate mapping from the image to the atlas, the
label for each image voxel can be determined by looking up the structure at the
corresponding location in the atlas under that mapping. Obviously, computing
the coordinate mapping between the image and the atlas is the critical step in
any such method. This step will be discussed in some detail in section 11.3.
