Biomedical Engineering Reference
In-Depth Information
be performed on two-dimensional planar or three-dimensional volumetric
data. A simple example shown in Color Figure 11.2* is of an x-ray of the
hand coregistered with the corresponding bone scan. In this example the
bone scan was performed to detect sites of increased radiotracer accumula-
tion in the small bones of the hand, and to localize this as accurately as pos-
sible.
16
Small fractures may be present and detected in this way when they
are not apparent on x-ray alone. To perform such a registration, the patient's
hand is placed in a rigid frame for both the radiograph and the gamma cam-
era images. Fiducial markers, visible on both the x-ray and gamma camera,
are incorporated into the frame thus allowing the
realignment
and scaling of the data, which, for these purposes, are taken to be a rigid
object. The registration in this case is relatively straightforward: the operator
uses an interactive computer program to identify the corresponding mark-
ers in the two images and the registration algorithm determines the scaling,
translation, and rotation degrees of freedom to align the markers (see the
description of the Procrustes algorithm in Chapters 2 and 3 for more details
of this approach). Once registered, any bone which has abnormal uptake can
be accurately localized. Most registration tasks, however, are more compli-
cated than this example. The majority of registrations involve three-dimen-
sional volumetric data, often between different modalities. Many of the
early algorithms developed for spatial registration were intended for use
with different modalities, e.g., PET and MRI.
retrospective
17
Spatial registration provides an anatomical framework in which to inter-
pret the functional emission data arising from SPECT and PET. Combining
functional PET data with structural MRI data in neuroscience research was
one of the first applications for PET-MRI registration that achieved wide-
spread routine use. In a paper in 1993, Watson et al. demonstrated, for the first
time in man accurate spatial localization of the functional locus of an area of the
brain outside the primary visual cortex which was predominantly interested in
visual motion alone
18
known as “V5.” Given the limited spatial resolution of
the PET blood flow studies, it would have been difficult to localize, as pre-
cisely as was done in this paper, the exact location of the functional area V5
in the parieto-occipital cortex without the aid of accurately registered MRI.
The data were registered with an automated algorithm where the parameter
that was optimized, also known as
cost function
, was the partitioned image
17
uniformity (PIU) measure,
described in Chapter 3. Today it is standard to
represent areas of change in cerebral activity measured with PET or func-
tional MRI (fMRI) superimposed with a high resolution structural MRI scan.
This approach also has application in coregistering brain scans containing
lesions with, for example, -DG (PET) or [ ]-HmPAO (SPECT) scans.
The registration problem in the case of the human brain is simplified in that
it can be treated as a rigid body, and therefore a transformation allowing
only for six degrees of freedom (three translations and three rotations) plus
99m
18
[]
F
Tc
* Color Figures follow page 22.
