The application of iterative reconstruction algorithms to positron emission tomography (PET), in particular the expectation-maximization (EM) algorithm for maximum likelihood (ML) reconstruction, has been widely studied although not widely implemented for routine reconstruction of PET data. Efforts to date have focused on the fundamental issues related to the ML algorithm (e.g., stopping criteria, constraints to improve convergence, image artifacts). Previously, we studied methods of inclusion of the corrections for physical factors associated with PET data, namely, radial resolution nonuniformity, detector normalization, scatter, attenuation, and random coincidences, in an attempt to obtain quantitative two- dimensional (2-D) images with the ML algorithm. This year we extended the ML algorithm to three-dimensional (3-D), or volume, PET data. NIH is one of the few institutions with and PET scanner capable of acquiring volume PET data and the necessary computing power to carry out a full 3-D ML reconstruction. Even with the massively parallel computer used for the 2-D algorithm, however, the 3-D problem is unwieldy, in terms of the time, memory, and file sizes required. Implementation of the 3-D ML algorithm required care in order to take advantage of time and size reduction techniques. An initial study of the convergence of the 3-D ML algorithm has been performed; an investigation of the noise and bias properties of the algorithm is currently underway. The 3-D ML algorithm will be compared with analytic methods of 3-D reconstruction of PET data, both """"""""exact"""""""" and approximate, to assess the characteristics of images generated with the different algorithms. The final goals of this study are to determine the circumstances where the computationally-intensive ML algorithm may be superior to a faster, but potentially biased, analytic approach, and to investigate modifications to the 3-D ML algorithm which will improve the speed of reconstruction while retaining the advantages of the algorithm.
