Biomedical Engineering Reference
In-Depth Information
with fields of view providing sufficient coverage of the relevant part of the
body in both image sets. MR images are usually reconstructed to 256
256 pixel
matrices with pixel size of 1 mm or less and with variable slice thickness
between 1 and 10 mm, depending on the organ of investigation. Often matri-
ces of 512
512 pixels are available. In addition, the image resolution (1 mm
or less) which is primarily not hardware-dependent, may be optimized for a
specific task by selecting the most appropriate pulse sequence. For clinical
purposes often only a rather limited number of slices is acquired, so that the
total measurement time is kept short, partly with the aim of avoiding the risk
of motion artifacts. In these cases the slice thickness is chosen to be 4 to 6 mm
for brain studies and 4 to 8 mm for nonbrain studies. The volume effectively
covered by all images can be adjusted by varying the slice thickness or number
of slices acquired. Also, the image orientation can be selected by the operator
and adapted to obtain the best imaging result. Transverse, coronal, sagittal, and
even oblique slices can be obtained. For brain studies, transverse orientation is
often preferred, whereas for the thorax mostly coronal images are acquired.
Most of these properties are quite different in PET imaging: PET images are
typically reconstructed into matrices with 128
128 pixels of size of 2 to 4 mm
or even greater, with resolution of 4 to 8 mm, depending on the type of scan-
ner. The slice thickness and number of slices are generally defined by the
hardware. Also, the primary image orientation is restricted to transverse slices.
The axial field of view of modern PET scanners is rather limited (16 to 25 cm),
and so most of the systems offer the option to acquire a series of scans, each
at a different offset of the patient bed relative to the scanner gantry. Studies
with multiple bed positions thus provide image sets that may even cover the
whole body, but are frequently limited to only 3 to 5 sequential positions cov-
ering up to 80 cm of the body.
The fact that PET and MR images contain different information and have
different resolution, representation, and orientation complicates their com-
parison and interpretation. If accurate spatial registration is to be performed,
pixel size and interslice distance have to be precisely known prior to starting
the registration process. (It should be noted that a large pixel size leads to
images prone to higher partial volume effect, i.e., averaging of signal intensi-
ties across a larger volume, leading to images with lower contrast.) It may
also be necessary to correct possible geometrical distortions introduced by
the imaging devices as described in Chapter 5. Another prerequisite is that
the primary orientation of the images should be similar. Consider the case of
coronal MR images with good in-plane resolution (pixel size 0.5 mm) but
with a large slice thickness (6 mm). MR images with these properties are cer-
tainly more difficult to register to PET images, which have nearly isotropic
resolution of 4 to 5 mm, than using a volumetric MR image set of 1 mm pixel
size in all dimensions, or transverse images instead of coronal. Since the reg-
istration procedure usually involves reslicing of an image set by interpola-
tion, the quality of the original coronal MR images would certainly be
degraded, which might cause problems for certain image registration algo-
rithms or for sufficient accuracy of the registration.
