Biomedical Engineering Reference
In-Depth Information
help in resolving difficulties in implementation of image registration. The
considerations of image acquisition and data preparation described here
are pertinent irrespective of the quality, calibration, state of repair, etc., of
the imaging hardware employed. In addition, it is also frequently necessary
to take account of errors introduced in images as a result of deficiencies
in the image acquisition process. These scanner errors are the subject of
Chapter 5.
Once the images have been registered and so share a common coordinate
system, they will usually need to be assessed or analyzed together, as well as
individually. Display of the images to facilitate comparison can present chal-
lenges, and a number of methods have been devised. Some common display
techniques are also reviewed.
4.2
Image Acquisition
In general, registration of data representing a three-dimensional (3D) object,
such as a brain, requires full 3D data coverage. However, many imaging
modalities produce individual 2D images analogous to slices through the
object, or projections of the 3D object onto a 2D plane (e.g., plain x-ray
images). When individual 2D image slices are acquired, these can readily be
registered with one another, but the information they contain may be irrecon-
cilably mismatched because of movement of critical structures out of the plane.
When registering two 2D projection images, satisfactory registration is often
impossible because structures overlapping in one view do not overlap in the
second. Modalities such as MRI, CT, SPECT, and PET are frequently used to
provide 3D coverage, but this is often in the form of discrete slices which may
only provide partial information in the through-slice direction. Such multislice
data may deliberately be obtained with gaps between slices to increase acqui-
sition efficiency and
or minimize cross talk between slices. When such data are
resliced to a new set of image planes, the missing information at the edges of
the original slices results in errors in the final data. Thus, the resliced images
are different from those that would have been obtained if a direct image
acquisition had been performed at the final slice location. The impact of these
errors depends critically on the application and on their magnitude and loca-
tion. For example, in multimodal registration, where CT or MR images are
being used to provide contextual anatomical information, errors in intensity
or incorrect resolution of fine structural detail may be of little or no conse-
quence. By contrast, in single modality serial studies using rigid-body regis-
tration, such as fMRI, small local intensity errors produced by reslicing
incomplete data may be as large as the effects being studied and so may mask
true effects or produce false positive results.
A key feature of errors introduced by reslicing incomplete data is that such
errors are local in the sense that individual pixels or groups of pixels get cor-
rupted, and the nature of the errors is related to the underlying structure and
