Art vs. Convolution Algorithms for ECT

Abstract

Iterative algorithms were popular in the early days of computed tomography, but fell out of favor because they were considerably slower than convolution algorithms, especially when implemented with the Fast Fourier Transform (FFT). Some of the early work appeared to indicate, however, that iterative methods have better noise handling capabilities than convolution methods (1). In an excellent overview of the iterative reconstruction algorithms Herman and Lent (2) noted that the ART-type were more accurate than the convolution algorithms when the reconstruction problem was underdetermined (too few measurements for the number of unknowns), the picture had high contrast, and there was a large amount of noise in the data. The latter two points describe the usual situation in nuclear medicine tomography, and the first point (an underdetermined problem) would hold if we desired, or were forced, to limit the number of angles at which data were collected.