Image Processing Techniques in Emission Tomography - Correction Techniques in Emission Tomography - page 120

images or an image description) with desired properties is generated by pro-
cessing one or more input images. Many common image processing techniques
from computer science are used in emission tomography, e.g., noise reduction
or image registration for motion correction.
The transfer of general image processing techniques to medical imaging is
usually straightforward but sometimes specific modifications are necessary or
beneficial. For instance, medical imaging is usually associated with 3D volumes
or even 4D time series, unlike 2D images in traditional image processing. Ad-
ditionally, the inclusion of prior knowledge, e.g., information about anatomy
or the acquisition process, can be advantageous.
The elds of application are manifold|and so are the image processing
algorithms. Accordingly, we can give only a non-exhaustive overview of basic
techniques commonly used in emission tomography. This chapter is mainly
aimed at those not familiar with image processing and may serve as a starting
point for further studies.
Noise removal is an important preprocessing step for many tasks in image
processing|especially in medical image analysis. In Section 7.2 we describe
common methods for noise removal and distinguish between image, Fourier,
and wavelet transform domain based methods. By definition, image processing
denotes the manipulation of images. This manipulation often includes resam-
pling of the data, i.e., interpolation, as discussed in Section 7.3. Image regis-
tration provides the fundamentals for image alignment and motion detection.
Hence, we present a categorization of registration techniques in Section 7.4
where we put a particular focus on the meaning of \similarity." We con-
clude the section with a brief review of current registration software packages.
When approaching the limitations of scanner resolution, the partial volume ef-
fect (PVE) becomes relevant. An introduction to PVE and techniques dealing
with the correction of PVE are presented in Section 7.5. Super-resolution is
a promising approach to overcome resolution limitations of image acquisition
and is discussed in Section 7.6. A major problem in developing algorithms in
medical imaging is the lack of ground-truth data. This applies to a wide range
of algorithms, e.g., motion detection and segmentation. We discuss general
validation methods including soft- and hardware phantoms in Section 7.7.
All figures in this chapter were created using MATLAB r . For Wiener filter-
ing, Fourier and wavelet transform, and deconvolution the respective functions
provided by MATLAB r were used. PET images are shown using an inverted
grayscale colormap so that dark colors indicate high activity and light colors
low activity, respectively.