Biomedical Engineering Reference
In-Depth Information
and skin, the exact placement of the devices during the imaging session has
to be carefully maintained, because subsequent processing is likely to assume
a fixed relationship of the marker positions with respect to the patient's head
or brain. This can be problematic to achieve.
An advantage of using external markers with PET is that the markers are
directly visible and so do not depend on the activity distribution of a partic-
ular tracer. As pointed out previously, PET images may display abnormal
focal uptake with major anatomical landmarks much less pronounced.
However, external markers can introduce imaging artifacts in the reconstruc-
tion, especially with filtered backprojection, if they have a relatively high
radioactivity concentration compared to the activity of the normal tissue.
The application of external marker devices in MRI may be hampered by the
fact that these devices could interfere with the standard head coils necessary
for imaging. In addition, external markers are usually placed some distance from
the head, where possible distortions might be largest. In any case, the position of
the markers has to be accurately deduced from the images, requiring strict com-
pliance with the imaging protocol to be adopted. The field of view must be
large enough for the markers to be included in both scans, and the images
must have sufficient sampling in all three dimensions to accurately locate the
markers. This might be in conflict with the standard protocols, especially
with MRI, when the slice thickness is usually preset to a larger value (4 to
6 mm) for practical reasons. Registration algorithms based on point markers
are further discussed in Chapter 3, Section 3.4.1 and Chapter 6.
The second group of registration algorithms encompasses a variety of post-
acquisition techniques with less stringent requirements in patient handling.
These retrospective techniques, in general, rely solely on the information con-
tent within the images, with each pixel representing an intensity value or a
physiological parameter at each particular location inside the image volume.
Although both measurements (with PET and MRI) are performed indepen-
dently (i.e., no special hardware enforces identical patient positioning), it is
advisable to design the acquisition protocol appropriately in order to ease the
subsequent registration step. This includes the proper selection of primary
image orientation, pixel size, and slice thickness, whenever this is at the oper-
ator's discretion (see also Section 9.4.1 for further details).
Two subgroups of retrospective algorithms can be identified. One com-
prises techniques which perform the registration in an automated manner
with the optimization step to find the best transformation based on similarity
measures described in the literature (see Studholme et al.
10
for a recent review
and discussion, and Chapters 2 and 3). Utilizing a similarity measure assumes
that the images to be registered bear sufficient similarity to each other. A simi-
larity can be based on a wide variety of properties, which need not be linearly
correlated. Gray matter structures with medium intensity values on T1-
weighted MR images are bound to correlate with areas of higher uptake in
FDG PET images, while white matter structures (with higher intensity values
on this type of MRI) correlate with relatively lower values in the PET images.
This relation does not hold any longer when other tracers with a different
