Signal Processing - May 2016 - 75

full-width at half-maximum. Any object with smaller dimensions only partly occupies this characteristic volume, such
that acquired counts are spread over a larger volume owing to
the limited spatial resolution of the PET scanner. Under these
circumstances, corresponding PET images still reproduce the
total amount of radiotracer uptake within an object but do not
represent the regional activity distribution within this volume.
A number of strategies have been proposed to correct for
PVE [60]. The most straightforward approach uses recovery
coefficients, which can be determined through experimental
studies involving the use of spheres of different sizes. This
simple approach produces acceptable results for objects with
similar shape as the calibration phantom used for derivation
of recovery coefficients (e.g., tumors of spherical shape). More
refined approaches rely on anatomically guided postreconstruction techniques, in which the size and shape of corresponding objects assessed by structural imaging (MRI or CT)
[61] are used instead to correct for this effect.
The PVE is among the major concerns in brain PET imaging in connection with quantification of cerebral metabolism in
the atrophied brain, such as with Alzheimer's disease. Various
voxel-based MRI-guided PVE correction methods have been
proposed. The most popular technique consists of segmenting
MR images into white and gray matter after PET/MRI registration. This is followed by convolving the segmented white
and gray matter images by a Gaussian point spread function
representing the PET scanner's spatial resolution. The PVEcorrected gray matter PET image is achieved by subtracting
the convolved PET white matter image from the original PET
image, followed by division by the convolved gray matter MR
image. The final step involves the application of a binary mask
to the gray matter region [62].
The overall accuracy achieved by MRI-guided PVE correction in PET depends upon the accuracy achieved by each
procedural step, including image registration and MRI segmentation. This has been investigated in detail for the voxelbased approach [61]. The high soft-tissue contrast provided by
MRI provides reasonable accuracy in terms of differentiation
between gray and white matter. Nevertheless, errors in segmentation of brain tissue components have been found to be of
greater significance [63]. For instance, a 25% error in total volume produces a 5% decrease in the caudate nucleus apparent
recovery coefficient [64]. It is interesting to note that the effect
of segmentation error is limited to the missegmented region.
Inaccuracies from segmentation can be regarded in the framework of a more broad question of tissue heterogeneity. In fact,
the main limiting feature of these algorithms is the assumption
regarding the homogeneity of radiotracer distribution in each
region or tissue component. Overall, it appears that the success
of MRI segmentation has a higher impact on the accuracy of
the corrected estimates [63] compared with the influence of
image registration, although some studies seem to suggest that
registration errors have the greatest impact on data accuracy
and precision [61].
More refined strategies using multiresolution synergetic
approaches merging anatomical and functional information

seem to have the potential to overcome the limitations of classical techniques. However, their feasibility in a clinical setting
still needs to be demonstrated [65]. PVE correction can also
be included directly into statistical reconstruction algorithms
through the use of an appropriate mathematical formulation of
PVE in the forward model along with other physical degrading
factors governing the physics of PET [58].

MRI-guided motion compensation in PET/MRI
Recent advances in PET instrumentation have made it possible to achieve high spatial resolution, which motivates further development and clinical implementation of sophisticated
motion correction strategies. The various sources of motion,
including unwanted patient motion, cardiac motion, and respiratory motion, and correction strategies specifically developed
to reduce or eliminate them have been reviewed recently [66].
Overall, three broad approaches were reported in the literature:
1) nonrigid registration of independently reconstructed images;
2) initial estimation of motion information from gated PET or
MR/CT images, subsequently used in a new reconstruction
applied to all gated frames; and 3) simultaneous estimation of
motion parameters and images.
Motion between or during anatomical/molecular data
acquisition remains an important challenge for PET/MRI
protocols. The characteristic misalignment between PET and
CT images at the level of the diaphragm in PET/CT systems
resulting from breathing pattern differences is expected to be
partly addressed by PET/MRI owing to the longer acquisition time of MRI sequences used for attenuation correction,
which results in temporal averaging that would improve PET
and MRI registration in some situations. In addition, the use
of a specific respiratory protocol in PET/MRI can improve
the spatial correspondence between PET and MRI. Owing to
the typical duration of PET data acquisition (2-3 minutes/bed
position), a PET image corresponds to an average of several
respiratory cycles and is susceptible to motion-related distortion. Similarly, typical low-resolution MR images suitable for
attenuation correction involve averages over multiple respiratory cycles, although the averaging process in MRI is different
from that in PET. More importantly, severe motion artifacts
may appear when there is marked organ motion with increased
noise and smaller-appearing organ size on the MRI attenuation map, with subsequent bias in the attenuation correction
procedure. Ideally, PET and MR images should correspond
to the same phase of the respiratory cycle and be matched to
achieve accurate attenuation correction and improved spatial
resolution. To achieve good matching between PET and MR
images at a specific respiratory phase, the patient's breathing
during scanning should be synchronized to reduce distortional
effects of respiratory motion. Provision of breathing instructions to patients prior to scanning may also be useful.
An assortment of MRI motion-tracking methods predominantly for rigid-body motion have been employed in the clinical
setting, including, but not limited to, embedded cloverleaf navigators [67]. One such technique uses motion estimates derived
from high temporal resolution MRI during simultaneous

IEEE Signal Processing Magazine

May 2016

Table of Contents for the Digital Edition of Signal Processing - May 2016