Biomedical Engineering Reference
In-Depth Information
between respective image spaces alone, without reference to physical space,
the degree of spatial integrity of each dataset used for registration taken indi-
vidually is an important consideration in all cases. For example, in ET it is
often desirable, even essential, to place the reconstructed functional data
within an “anatomical framework.” Although image distortion in ET is not a
primary concern when registering with CT or MRI, a number of issues
related to image quality can affect image registration and therefore require
careful consideration. Another example is in longitudinal MRI studies, which
can also be impaired by variations in image characteristics (such as voxel
dimension), although mapping between image space and real space may not
be sought. In the following sections, we begin by considering the nature and
origin of artifacts in medical images, as this will allow us to define the scope
of the problem. We then review specific relevant artifacts and describe the
current state of the technology available to correct for these artifacts for each
imaging modality.
5.1.1
Image Artifacts, Geometric Distortions, and Their Origins
The ideal image acquisition process results in a representation of the imaged
object devoid of artifacts and noise. In other words, pixel intensities in the
ideal image have a spatial distribution that matches exactly, to within a global
dimensional scaling factor, that of the physical property(ies) of the object
imaged. Artifacts can therefore be seen as occurrences of signal intensities
that violate that correspondence. In the context of image registration, all arti-
facts are undesirable and can impair the registration process. However, the
most relevant artifacts are image features that violate the requirement for
an appropriately scaled, Cartesian mapping between image space and real
space,* i.e., geometric distortions. Artifacts that appear as features superim-
posed onto the image, such as coherent noise or signal nonuniformity (e.g.,
due to radiofrequency inhomogeneity in MRI), must also be considered, as
they may impede registration, although they do not represent geometric
distortions. No further discussion of random image noise is necessary, as it is
intrinsic to any data acquisition process, can originate in the imaged object or
imaging device and, therefore, cannot be classified as scanner error. These
considerations lead to discussion of the origin of image artifacts. Image arti-
facts can be categorized according to the location of their origin (either
imaged object or imaging device) and the generating mechanism. The possi-
ble mechanisms of artifact generation are multiple, but can be categorized as
either intrinsic or resulting from interactions between imaged object, envi-
ronment, and imaging device. Mechanisms of artifact generation intrinsic to
the imaged object include spontaneous movement and changes in its prop-
erties (e.g., temperature, mechanical properties, etc.). Mechanisms intrinsic
* For 3D imaging modalities. For projection methods, the mapping most often assumed is that
of a point perspective.
