Image Processing Techniques in Emission Tomography - Correction Techniques in Emission Tomography - page 130

(a) -map (b) PET (c) Overlay

FIGURE 7.9: The -map (a) was registered to the PET image (b) for atten-
uation correction. (c) Overlay of the PET image and the -map. The colormap
for the -map is chosen similarly to the colormap of the PET data (inverted
grayscale) for better visibility.

PET is primarily the correction for absorption of photons in the body. This
is done by registering an attenuation map (-map) to the PET data and
incorporating this registered attenuation map into a second reconstruction
of the PET data. The -map is usually obtained from a separate CT scan.
In Figure 7.9(a) the -map registered to the PET image in Figure 7.9(b) is
shown. An overlay of the -map and the PET image is shown in Figure 7.9(c)
to visualize the accuracy of the alignment.
As seen in this introductory example, registration in emission tomography
is inevitable due to several tasks requiring spatial alignment of correspond-
ing images. Especially in combined approaches like PET/CT or PET/MRI
an optimal alignment is mandatory as information of both techniques (func-
tional and morphological) is combined. Also, there are numerous examples of
monomodal registration tasks for scans taken at different times (e.g., tumor
growth studies, motion correction in gated PET (see Chapter 8), rest-stress
comparisons) or of different objects (e.g., normalization to atlas data).
In the following, image registration algorithms are categorized in Sec-
tion 7.4.1. In Section 7.4.2 methods for validating registration results are
discussed. For practical use some freely available registration software is pre-
sented in Section 7.4.3. For sustained discussion beyond the scope of this gen-
eral overview we refer to some comprehensive survey articles about registration
in general [4, 9, 13, 72] and medical imaging in particular [23, 27, 37, 41, 55, 61].

7.4.1 Categorization
Throughout the rest of this section the following notation will be used.
The image to be transformed is the template image T and the static image is
the reference image R. Let d denote the spatial dimension and R d the
domain. Then the template image T and the reference image R are defined
as T;R : !R.
To get an overview of the huge variety of different registration approaches,
a categorization using criteria adopted from [41] follows. In this context the
nature of transformations and similarity measures are discussed in more detail
in Sections 7.4.1.1 and 7.4.1.2.