Comparison Of Bayesian Methods For Sinogram Filtered Backprojection And Algebraic Reconstruction

Abstract

Image reconstruction by Maximum a Posteriori sinogram restoration and Radon inversion (MAP-RI) was evaluated. Its equivalence with Maximum a Posteriori al- gebraic reconstruction (MAP-ART) was established. A fast practical implementation of MAP-RI was developed using truncated priors and filtered backprojection (MAP-FBP). Simulations using one plane of the Hoffman 3-D brain phan- tom at both low SNR and high SNR were used for quanti- tative comparisons of resolution and noise variance. Com- pared to conventional FBP with ramp cutoff at Nyquist, MAP-ART can produce images with 40% greater resolu- tion and comparable noise variance in the high activity re- gions of phantoms. In comparison, MAP-FBP can yield a 30% improvement in resolution with equivalent noise to FBP. However, MAP-FBP using an iterative expectation- maximization (EM) algorithm converges rapidly (< 10 iter- ations) with an order-of-magnitude lower cost per iteration compared to MAP-ART.