INTRODUCTION - Fundamentals of Three-Dimensional Digital Image Processing - page 16

Table 1.1. Examples of meanings of density values in 3D images.

attenuation coe cient of X-ray at a small volume
of an object

X-ray CT images

magnetic resonance images strength of resonance signals at a small volume

element of an object

ultrasound images absorption, reflection, or transmission coecient

of ultrasound wave at a small volume element

confocal laser microscopi images strength of reflected light or distance to the
surface of an object

positron emission CT (PET) absorption rate of γ -ray at a small volume of

an object

(1) Applying some processing methodology to a set of independent 2D images
and using the result to build or reconstruct the 3D image ( 2D image
processing with 3D image reconstruction )
(2) Processing each cross section, but during processing taking into considera-
tion the status and consistency of neighboring slices (for example, vertical
relationships), and using such processing methods as necessary. (Here, this
will be referred to as 2.5D processing )
(3) Performing 3D processing on 3D image data ( 3D processing ).

The decision also will be dependent upon the characteristics of the input
image. When the sampling distance between cross sections is large as com-
pared to the resolution (size of pixels) within cross sections, if additional cross
sections will not be inserted, then 2.5D processing may be effective. On the
other hand, if a true 3D image with approximately similar resolutions is ob-
tained, the 3D processing should be used whenever possible. Conversely, as
humans excel at intuitive judgments related to 2D images, starting out with
the 2D image processing with 3D image reconstruction discussed in (1) above
may not be without merit.
The focus of this topic is 3D image processing. If a true 3D image of the
subject is available, then the 3D processing of (3) above should be performed
whenever possible. However, in actual practice the other two methods also
should be put to use as circumstances demand. One must also keep in mind
that the condition of the input image or the computing environment may force
such decisions.
Let's look at an example. Figure. 1.4 shows a schematic image containing
an arboroid structure like that of a bronchial tube. Here, ellipsoids might be
extracted from 2D cross-sectional images of the tube using 2D image pro-
cessing, and the extracted ellipsoid structures smoothly connected to extract