Biomedical Engineering Reference
In-Depth Information
9.4.2
Techniques and Procedures Used in Clinical Applications
During the course of developing solutions to the problem of image registration
many techniques have been presented, some of which are suitable for applica-
tion only in very specific situations, i.e., connected to specific hardware, imple-
mented in an exotic computing environment, or requiring very high
performance computer systems. Hence, not all techniques developed to regis-
ter medical images are currently optimal for usage in a clinical setting. If image
registration is to be applied on a routine basis during the diagnostic decision-
making process, the techniques employed have to be robust against imaging
artifacts, reliable, fast, easy to use, and at least partly automated, i.e., avoiding
too much user interaction and with little or no preprocessing. As will be dis-
cussed later, the technique to be chosen depends on the specific application.
The main distinction concerning the techniques applied in a clinically ori-
ented environment is again between handling brain and extracranial images.
Furthermore, it is the specific application that defines the demand on precision
and accuracy. For example, radiation therapy planning requires different meth-
ods than neuroscience research and development applications. Other aspects
of presurgical diagnostics need a much higher accuracy than image correlation
in the thorax would reasonably ask for. The former application requires an
accuracy of about 1 mm or less. Based on the results from the registration step,
a stereotactic intervention (the placing of a radioactive implant) is performed
or the site and path of a biopsy needle is defined. A failure to accurately register
the respective studies could cause unwanted damage of healthy tissue or vas-
cular structures. In the thorax, the circumstances of image acquisition in sep-
arate scanners, using different patient beds with a different shape, and the risk
of respiratory artifacts make it much more difficult to achieve an accuracy of
1 mm or less.
From a clinical point of view the techniques useful for MRI-PET registration
can be classified into two main groups.
7
Some of these have been applied also to
work with other types of tomographic modalities, whereas other registration
techniques are dedicated developments for the registration of PET to MRI. The
intention of most of these developments has been to register brain images. Unless
otherwise stated, the following explanations will assume these types of images.
The first group of registration techniques represents prospective proce-
dures designed with the intention of performing both studies (PET and MRI)
following a strict dedicated protocol. An example is the application of a ster-
eotactic head frame
8
prepared with appropriate markers visible in both
modalities, and designed from materials like wood, in order to be applicable
for both PET and MRI. A variety of head holders or face masks have been
considered to establish a common coordinate system for both studies (e.g.,
Bettinardi et al.
9
). These prospective techniques focus mainly on brain imag-
ing, although there have also been attempts to develop similar utilities for
nonbrain imaging. However, these approaches tend to be demanding on both
subject and operator. Depending on the patient, a face mask might not be
well tolerated. Since face masks are mounted externally to the patient's head
