Biomedical Engineering Reference
In-Depth Information
imaging modalities are subject to the same sort of problem, but to a
much lesser extent in the case of CT and MRI.
I
MAGE
R
EGISTRATION
Both Figure 3.14 and Figure 3.15, above, show side-by-side CT and
MR images and, for these cases, such a presentation is adequate
to spatially correlate the two studies and to extract the clinical
information of interest. However, for more subtle comparisons one
wants to be able to accurately correlate points in one image with
points in the other so as to be able to combine the information from
two studies in a spatially accurate manner.
7
This process is referred
to as image registration or, alternatively, as image fusion. This topic
has been been reviewed by,
inter alia
, Maintz and Viergever (1998)
and by Kessler (2006). I first discuss image registration assuming the
two images data sets are spatially accurate representations of a rigid
body, and then for the situation in which one or both of the images are
spatially distorted, or in which the patient has changed shape between
the two studies.
Rigid body image registration
Consider first the inter-registration of
two 2D sections such as a pair of
transverse slices, either a CT and MRI
section, or two CT sections, or a
radiograph and a DRR. In the case of
two 2D images, one might think that
one could simply superimpose two
semitransparent representations of the
images upon one another and slide
one over the other until they
“matched.” This was, indeed, the first
thing I tried, many years ago.
However, this does not in general
Figure 3.17. The problem of
matching two disparate
studies. Figure courtesy of
M. Kessler, Madison, USA
.
7
Mathematically, one needs to compute the transformation matrix of
translations and rotations (4 degrees of freedom in 2D, 6 in 3D) and,
potentially, scale factors in each dimension. Of course, there may be more
than two studies of interest. In that case, one can inter-register two studies,
then additional studies of interest can be pair-wise inter-registered and
hence any two studies can be related to one another.
