Biomedical Engineering Reference
In-Depth Information
development of more sophisticated registration techniques leading to the
introduction of frameless stereotaxy in the mid 1980s,
11
though it was another
decade before frameless stereotactic systems obtained the regulatory
approval necessary for widespread use in health care.
In image-guided interventions, correspondence is established between
image and the physical space of the patient during the intervention. Estab-
lishing this correspondence allows the image to be used to guide, direct, and
monitor therapy, akin to providing a 3D map for navigation, with the aim of
making the intervention more accurate, safer, and less invasive for the patient.
In the last few years, image registration techniques have entered routine clinical
use in image-guided neurosurgery systems and computer-assisted orthopedic
surgery. Systems incorporating image registration are sold by a number of
manufacturers.
Stereotactic frames can also be used for intermodality image registration, but
their use is restricted to highly invasive surgical procedures because of the
need for rigid fixation to the skull. Various relocatable frames were proposed
to avoid this invasiveness, but beginning in the mid 1980s, registration algo-
rithms were devised that were “retrospective,” that is, did not require special
measures to be taken during image acquisition in order for registration to be pos-
sible. Various approaches were introduced in the mid 1980s.
12-14
These tech-
niques were devised to make it possible to combine images of the same patient
taken with different modalities and they required substantial user interaction.
Another major step forward in image registration came in the first half of
the 1990s with the development of retrospective registration algorithms that
were fully or virtually fully automated for both intramodality
15,16
and
17-21
intermodality registration.
A significant breakthrough in the mid 1990s
was the development of image alignment measures and registration algo-
rithms based on entropy and, in particular, mutual information—measures
first derived from the information theory developed by Shannon in 1948.
22
Recently the focus of research in medical image registration has returned to
intramodality rather than intermodality registration, and to extending regis-
tration algorithms to handle the more complicated transformations needed
to model soft tissue deformation and intersubject registration.
Detailed atlases or computer models of anatomy are becoming available, in
particular from high resolution sources such as the Visible Human datasets.
4
The Montreal Brain Atlas has been generated by averaging images of the
brain across a population. Establishing spatial correspondence between these
atlases and an individual's images allows for easier interpretation and, in
particular, enables computer assistance in delineation of anatomical struc-
tures of interest.
Rapid advances in the power of computer technology and in the perfor-
mance of new registration algorithms and displays mean that image manip-
ulation deemed impractical or far too computationally expensive only a few
years ago can now be undertaken on the PCs available on most people's
desks. In our experience, initial work on voxel similarity measures was prov-
ing successful in the laboratory in 1994 but took an hour or more to complete.
