Corrections for Physical Factors - Correction Techniques in Emission Tomography - page 94

FIGURE 5.22: Fitting of 3D PET projection data (dashed line) with a
Gaussian (solid line) to determine the scatter distribution in a scan of a line
source inside a uniform cylinder.
5.5.3 Direct calculation methods
These methods are characterized by the usage of both emission and trans-
mission data as well as a scanner model to calculate the scatter distribution
according to basic Compton scatter physics (Klein{Nishina formula for the
electron-photon cross section [47]). These methods are numerically more com-
plex than the approaches described before; however, due to increasing compu-
tational speed available, they have become more and more important during
the last years.
Most of these methods aim to determine the single scatter distribution
(i.e., events where only one photon is scattered) as this is assumed to be the
most prominent part of the overall scatter distribution (single scatter simula-
tion, SSS) [72] [85]. Either appropriate scaling or integral transformations of
this single scatter distribution is then taken as a good approximation of mul-
tiple scatter events. More recent algorithms even allow highly accurate direct
determination of multiple scatter distributions [62]. Scatter from outside the
field of view can be a problem in these methods; however, algorithms that take
this explicitly into account have been developed and successfully implemented
[1].
Monte Carlo simulation techniques have also been used to calculate the
scatter distribution in emission tomography scans. In these approaches, an
estimate of the actual tracer distribution is used in addition to a model of
both the object and the scanner geometry to follow the fate of photons origi-
nating from voxels inside the object according to the tracer distribution [55].
A disadvantage of this approach is that scatter from outside the field of view
is not taken into account. Additionally, the actual tracer distribution neces-
sary for accurate scatter determination is not known, usually leading to a bias