Biomedical Engineering Reference
In-Depth Information
18
18
F]-fluoro-2-deoxy-D-glucose) on PET and
therefore is unlikely to represent recurrent tumor. Figure 2.7 shows a
sequence of CT axial slices, with the corresponding aligned
of high uptake of
FDG (2-[
18
FDG PET
images overlaid in pale green, taken through the pelvis of a patient who had
received previous radiotherapy for cervical carcinoma. The images clearly
show increased uptake in the denser mass shown on CT. This is likely to rep-
resent recurrent tumor rather than radiation-induced fibrotic changes, and
this was confirmed at surgery. Figure 2.5 and Color Figures 2.6 and 2.7* were
aligned using manually identified landmarks assuming that the part of the
patient imaged could be represented as a rigid body. Now this process is
almost completely replaced by the fully automated registration method
based on voxel similarity and described in Section 2.4.3.
2.4.2
Surface-Based Registration
Corresponding surfaces may be identified and used for registration. In these
algorithms, corresponding surfaces are delineated in the two imaging
modalities and a transformation computed that minimizes some measure of
distance between the two surfaces.
12-15
At registration this measure should be
minimum. The first widely used method was the “head and hat” algorithm,
13
15
but most methods are now based on the iterative closest point algorithm.
2.4.2.1
The “Head and Hat” Algorithm
In the “head and hat” algorithm, the contours of the surface are drawn on a
series of slices from one modality. This is called the head. A set of points that cor-
respond to the same surface in the other modality are identified. This set is
called the hat. The computer then attempts a series of trial fits of the hat
points on the head contours. The process of progressively refining these trial
fits is known as
. At each iteration the sum of the squares of the dis-
tances between each hat point and the head is calculated, and the process con-
tinues until this value is minimized. The hat now fits on the head. As its name
implies, this was first used on images of the head and, in particular, the align-
ment of MR and PET images. Unfortunately, just as there are many ways of
placing a real hat on a head, this algorithm can be prone to choosing the
wrong solution. These types of algorithms tend to fail when the surfaces
show symmetries to rotation, which is often the case for many anatomical
structures. The head can be rotated cranio-caudally (nodding) with minimal
displacement of the skin surface in a direction perpendicular to the surface.
This problem is illustrated diagrammatically in 2D in Figure 2.8. Figure 2.9
provides orthogonal cuts of an MR image and an overlaid PET image of the
head that are grossly misregistered, yet the surfaces of the brain are surpris-
ingly well aligned over much of the volume.
iteration
* Color Figures follow page 22.
