Biomedical Engineering Reference
In-Depth Information
or radiofrequency coils) or to their scattering out of the detector field of view (FOV)
is classified as
attenuation
. Attenuation, scattering and random events are causes of
noise and loss of image quality in PET images, with a consequent loss of precision and
detail, which means a decrease of clinical and diagnostic utility of the reconstructed
images. Therefore, in order to improve image quality, several techniques have been
developed to correct for these sources of errors and artifacts.
One possible approach to correct for random events is to add a parallel coincidence
circuit to the one measuring the prompt coincidences. Subsequently, the logic pulse
from one of the two detectors in coincidence is delayed in time, such that the detector
pair cannot produce any true coincidence. Therefore, detected events will be clas-
sified as accidental coincidences. It is then possible to extract the number of true
coincidences simply by subtracting the number of accidental coincidences from the
total number of detected events. This method is virtually free of systematic errors,
because the delayed and the prompt coincidences are measured in the same circuitry.
However, this subtraction leads to an increase in the statistical uncertainty [
37
].
The most widely used method for scatter correction is the analytic single scatter
simulation (SSS), which has shown a good reliability for whole-body PET [
38
,
39
].
Polycarpou et al. [
40
] evaluated the accuracy of such a method when applied to cases
with high scatter. There are two possible ways to apply attenuation correction (AC) to
the acquired data. One way is to compute the so-called attenuation correction factors
(ACFs), one for each LOR, store them in an attenuationmap and use such information
to pre-correct the acquired emission data before image reconstruction. The alternative
method is to incorporate the knowledge about the attenuation coefficients directly
into the iterative image reconstruction procedure. Earliest AC techniques performed
in stand-alone PET machines involved the use of rotating sources placed within the
PET gantry. In PET/CT devices, this early attenuation correction technique has been
replaced by CT-guided AC, based on the use of CT data to determine the attenuation
map containing attenuation coefficients for each voxel. ACFs are then computed by
integrating over each LOR. Currently, the challenge is to develop an efficient MRI-
guided AC for modern PET/MRI hybrid machines (Fig.
3.1
). The major problem
when facing MRI-guided AC, i.e. the preparation of an attenuation map, is that MRI
images do not provide any direct way of measuring tissue attenuation coefficients.
One method adopted to implement MRI-guided AC is to segment MR images into
different tissue classes and assign a pre-defined average tissue attenuation coefficient
to each class.
Segmentation-based AC
methods have been proposed for brain [
41
-
45
]
as well as for whole-body PET [
46
-
49
]. The idea of segmenting the attenuation map
into several classes of tissue was proposed for the first time by Huang et al. in 1981
[
50
] with the aim to reduce noise propagation from the acquired transmission images
[
51
]. The main problems connected with this technique are:
•
The segmentation of bones that appear black inMR images because of the short T2
relaxation times in MR data acquisition, making them difficult to distinguish from
air, since both air and bone tissue produce no signal inMR. On the other hand, bone
and air have opposite photon attenuation properties: bone is the highly attenuating
