AN INVESTIGATION OF GRADIENT-BASED RECONSTRUCTION ALGORITHMS FOR STATISTICAL PET TRANSMISSION IMAGING

摘要

For PET transmission imaging, the conventional iterative algorithms based on expectation maximization type algorithms, could not effectively converge to optimal image solution. In this study, we suggest a statistical model PET transmission data, and then investigate a class of gradient-based optimization algorithms for transmission image reconstruction including steepest ascent, conjugate gradient, and preconditioned conjugate gradient. From phantom studies, the preconditioned conjugate algorithms can converge to good image results within limited number of iteration. Combined with the suggested statistical model of transmission data, the preconditioned conjugate algorithms can also produce attenuation maps with accurate linear attenuation coefficients for clinical data.