3-D MAXIMUM A POSTERIORI ESTIMATION FOR SINGLE-PHOTON EMISSION COMPUTED-TOMOGRAPHY ON MASSIVELY-PARALLEL COMPUTERS

Since the introduction by Shepp and Vardi [1] of the expectation-maxim ization (EM) algorithm for generating maximum likelihood (ML) and maxi mum a posteriori (MAP) estimates in emission tomography, there have be en many investigators applying the ML method. However, almost all of t he previous work has been restricted to 2-dimensional reconstructions. The major focus and contribution of this paper is to demonstrate a fu lly 3-dimensional implementation of the MAP method for single photon e mission computed tomography (SPECT). The 3-dimensional reconstruction exhibits a major increase in resolution when compared to the generatio n of the series of separate 2-dimensional slice reconstructions. As ha s been noted, the iterative EM algorithm for 2-dimensional reconstruct ion is highly computational; the 3-dimensional algorithm is far worse. To accommodate the computational complexity, we have extended our pre vious work in the 2-dimensional arena [2]-[4] and demonstrate an imple mentation on the class of massively parallel processors of the 3-dimen sional algorithm. Using a 16 000 (4000) processor MasPar/DECmpp-Sx mac hine, the algorithm is demonstrated to execute at 2.5 (7.8) sec/EM-ite ration for the entire 64x64x64 cube of % planar measurements obtained from the Siemens Orbiter rotating camera operating in the high resolut ion mode.