Bayesian reconstruction of PET images: methodology and performance analysis.

摘要: We describe a practical statistical methodology for the reconstruction of PET images. Our approach is based on Bayesian formulation imaging problem. The data are modelled as independent Poisson random variables and image using Markov field smoothing prior. sequence calibration procedures which performed before reconstruction: (i) calculation accurate attenuation correction factors from re-projected reconstructions transmission image; (ii) estimation mean randoms component in data; (iii) computation scatter Klein - Nishina-based method. estimate then reconstructed pre-conditioned conjugate gradient quantitation study with multi-compartment chest phantom Siemens/CTI ECAT931 system. Using 40 1 min frames, we computed ensemble variance over several regions interest images standard filtered backprojection (FBP) protocol. values region were compared well counter each compartment. These results show that protocol can produce substantial improvements relative FBP protocol, particularly when short scans used. An example showing application method to clinical also given.