Biomedical Engineering Reference
In-Depth Information
an emerging trend towards accepting image registration for clinical conditions.
In the scientific literature there exist different approaches both to combining
images and to the interpretation of what is technically satisfactory. Therefore
some preliminary definitions are required.
The terms
of images are often used in the sense
that the images have entered a combined analysis. This could mean that both
studies were simply available either on radiological film or digitally on a
computer network, but were not manipulated by any computer program.
The physician mentally has to fuse the images and draw conclusions based
on the visual interpretation of the signals evident in the images. The terms
combination
and
correlation
registration
are generally used to
signify that data sets were processed to obtain images that correspond spa-
tially, i.e., on a pixel-by-pixel basis. Technically, this is most frequently solved
by applying computerized algorithms and, thus, it requires the images to be
readily available in digital form. Once the images are aligned, they can be
fused into a single display by various techniques (overlay, checkerboard
fusion, etc. — see Chapter 4).
,
matching
,
alignment,
and (sometimes)
fusion
9.5.2
Registration of PET and MR Images of the Brain
To date PET-MR image registration has been used mostly for brain images.
Because of the lack of anatomical details in the images provided by the early
PET systems, which had very poor spatial resolution, any aid from morpho-
logical images in defining the underlying anatomy was welcome. Image reg-
istration has entered clinical application in quite specialized areas such as
presurgical workup of brain tumors, in computerized treatment planning for
biopsies, open surgery, or radiation therapy, and in follow-up studies after
surgery or treatment. Although not performed for strict clinical purposes,
mapping of human brain function as visualized by O-15-water PET definitely
required registration of PET and MR images. In fact, such activation studies
may be part of the presurgical diagnosis of a patient with a tumor, for
instance, in the neighborhood of areas responsible for language generation.
14
Figure 9.6 shows an example for MRI-PET registration employing a combina-
tion of retrospective techniques (interactive and automated).
9.5.3
Registration of Extracranial PET and MR Images
While the task of defining an accurate and robust protocol for registration
of brain images appears relatively straightforward, the situation for other
images seems to be more difficult. Usually a rigid-body transformation is
the simplest approach, but certainly not the best and most accurate for
body images, where exact repositioning of patients in the respective scan-
ners is limited and motion caused by the heart cycle and by breathing may
complicate the application of algorithms successfully introduced for brain
images. Yet, it has been frequently noted in the literature that combining the
signals from both PET and MRI may improve the interpretation of the findings
and increase sensitivity and specificity in comparison to analyzing the images
