AbstractID: 8058 Title: Noise properties of hybrid fanbeam reconstruction algorithms: analytic
derivations and consequences
In this work, we derive and study an analytic expression
for the noise properties of a hybrid fanbeam reconstruction
algorithm we have recently developed. This hybrid algorithm
involves application of a Fourier-based fanbeam to parallel-beam
rebinning strategy followed by reconstruction by standard
filtered backprojection (FBP). The goal of the approach
was to avoid the distance-dependent backprojection factor
necessary in the standard fanbeam FBP (FFBP) algorithm, which
has been shown to lead to noise amplification in the periphery
of images, especially for small focal lengths. Such nonuniform
noise properties can affect the detectability of subtle lesions
and compromise the performance of image processing and analysis
algorithms.
While the hybrid algorithm leads to considerably more uniform
noise levels than does FFBP, the analytic noise expression
derived in this work reveals some low-grade structured nonuniformities
that are quite interesting. Specifically, the noise maps
contain a subtle ring pattern suggestive of the cross section
of a tree. The spacing and variation of the rings depends
on the focal length and is attributed to the radial linear
interpolation applied during rebinning. The implications
of these variations are considered, as are possible remedies,
including the application of more sophisticated interpolation
strategies to the radial rebinning step as well as approaches that
avoid interpolation altogether by reformulating the projection filtration
step to accommodate nonuniform sampling.