Biomedical Engineering Reference
In-Depth Information
In most modern volumetric images of the head acquired with a birdcage
head coil, the level of nonuniformity is such that it does not affect rigid-body
registration significantly, although it is still problematic for image segmenta-
tion.
38
However, images acquired with surface coils generally suffer from a
high degree of signal nonuniformity, which can hamper registration severely.
5.3.5.1
Correction
In the context of image registration, this problem can be addressed in two pos-
sible ways: first, by correcting images prior to coregistration and, second, by
using a registration method that is less sensitive to signal nonuniformity.
18
RF
nonuniformity correction methods can be classified into two main categories:
experimentally derived correction field maps and purely postprocessing. The
first approach is based on the assumption that the nonuniformity field is solely
dependent on geometry of the coil and therefore that a nonuniformity map
obtained by scanning a uniform test object can be applied to all images
acquired with the same coil.
37,39
This requires registration of the correction
matrix with the image, which usually relies on the image position and orienta-
tion information supplied by the scanner (geometric prescription). This
approach assumes that the effect is independent of the geometry and electrical
properties of the imaged object, which is generally not the case.
40
Postprocessing methods, on the other hand, usually rely on the assump-
tion that signal nonuniformity due to RF inhomogeneity is restricted to low
spatial frequencies.
41
A number of correction methods are available that
require expert supervision.
42,43
Other methods have been proposed that
form an integral part of segmentation algorithms.
44-46
This integrated
approach is potentially extremely powerful, if complete segmentation of
tissues within the region of interest is the desired outcome (otherwise the
segmentation represents an unnecessary computational burden). Also,
these methods currently either require expert supervision or are applicable
to a single part of the anatomy. Sled et al. proposed a fully automated, non-
parametric method that solely derives the intensity bias field and gives
good results for whole-head data, providing input data of sufficient quality
to allow subsequent precise segmentation of the brain.
47,48
Another nonpara-
metric automatic postprocessing method has been applied successfully to
standard brain images as well as images acquired using surface coils.
49
5.4
Spatial Inaccuracies in Emission Tomography
Emission tomography (ET) is a technique for externally measuring radioac-
tivity distributions
in cross section. In contrast to the other modali-
ties discussed in this chapter (apart from functional MRI), the data
produced by emission tomography are
in vivo
functional
images. These represent
