Wavelet domain Bayesian method for high noise level PET image reconstruction

In this paper, a new maximum a posterior(MAP) method for PET image reconstruction defined in wavelet domain is proposed. Compared to the conventional MAP methods with Markov Random Field (MRF) prior models, the proposed method, named WD-MAP method, has better performance in characterize both local and global feature of reconstructed image due to the wavelet transform. Wavelet packet decomposition strategy is applied to further improve the reconstruction quality. The convergence speed of WD-MAP method is accelerated by adopting conjugate gradient(CG) technique. Simulated experiment suggests that the proposed method offers competitive performance in PET image reconstruction.