Biomedical Engineering Reference
In-Depth Information
Computer algorithms for image segmentation are widely available and
range from simple intensity thresholding with or without manual interven-
tion to complex automated procedures that include a wide range of image
processing tools.
15
The segmented images may be stored as gray scale images
with pixels in regions to be excluded reset to zero, or as coded masks used in
conjunction with the gray scale images. Examples of coding schemes include
labeling individual pixels or regions with a one for inclusion and a zero for
exclusion, or with numerical values that are associated with designated tis-
sue types.
4.6
Displaying Images
Many computer programs exist for displaying images in general and, in
some cases, medical images in particular. When displaying medical images it
is important that the images are correctly labeled. Confusion about the patient's
identity or orientation (e.g., confusion over left and right in transverse sec-
tional images) or incorrect information about details of data acquisition (e.g.,
which nuclear medicine tracer or MR sequence was used) could lead to incor-
rect diagnosis or treatment.
In addition to these usual concerns, a new set of display challenges arises
when dealing with registered images. Once the images are in a single data
space, it is natural to want to display them in a coherent way and explore
spatial relations within the images. Many methods have been devised to
achieve this, and the choice depends very much on the application. The fol-
lowing discussion is intended to introduce some frequently used methods
and is not a comprehensive review of the topic. More detailed treatments can
be found in numerous texts, for example, Bankman.
18
A much used early technique was to display the images side by side on a
workstation and employ a linked cursor so that the act of pointing to a loca-
tion in one image automatically locates the corresponding location in the reg-
istered second image. This is useful for initial exploration and for locating a
few localized features. For more global comparisons, image fusion has been
used, where data from both images can be viewed as one. Simple methods for
doing this include:
1. Color overlay, in which the data sets each contribute to each pixel,
but in different colors, or one is coded in hue and the other in gray
scale intensity. If features from one image are segmented they may
be overlaid in isolation on another aligned image. For example,
blood vessels extracted from an angiographic acquisition could
simply be superimposed on an anatomical scan, so vessels can be
related to anatomical landmarks, even though at the site of the
vessels, none of the anatomic image can be seen (e.g., Chapter 10,
Color Figure 10.5).
