Regularized MLEM reconstruction with a strong anatomical prior using newton iterative algorithm

MAP algorithm has outperformed in medical image reconstruction with noise suppression and edge preservation. However MAP-OSL algorithm could hardly apply a strong prior by using a large regularization parameter. In this study we proposed a MAP-Newton reconstruction framework to enable a strong prior to be applied in MAP optimization. EM iterative framework was used in MAP-Newton. In the step of maximizing expectation of the complete data log-likelihood function, MAP-Newton solves the non-linear equation accurately with Newton iterative algorithm, which is potentially better than approximate linearization method used in MAP-OSL. MAP-Newton reconstruction algorithm was implemented with both Bowsher's prior and joint total variation (JTV) prior based on anatomical image. ¹⁸F PET and CT data of an image-quality phantom were acquired on the small animal PET/SPECT/ CT Iniview 3000 system for performance evaluation. The priors with three different strength (regularization parameter was from 0.01 to 1.0) were applied in the list mode reconstruction studies with three noise levels (100 M, 12 M and 2 M LORs separately). The normalized standard derivation (NSTD) of region of interest was calculated. The results with Bowsher's prior for all noise level cases showed no significant difference between MAP-Newton and MAP-OSL when the regularization parameter was smaller than 1.0. When the parameter was set to be 1.0, the proposed MAP-Newton can greatly reduce NSTD with reasonable image quality although no reasonable images could be obtained with MAP-OSL. The results with JTV prior showed the same trend, except that increased regularization parameter did not reduce NSTD. In conclusion, with a strong prior, MAP-Newton can result in better image quality than MAP-OSL in terms of noise suppression, while no significant difference when applying a light prior. It indicated that the MAP-Newton reconstruction with a strong prior could be applied, when we have enough confidence on the prior.