Comparison of noise performance for Penalized-Likelihood image reconstruction methods with uniform resolution estimators

Maximum Likelihood (ML) based reconstruction methods produce increasingly noisy images at higher iteration numbers. Penalized-Likelihood (PL) methods are used to overcome this noise where they use a penalty function in addition to likelihood function to impose constraints on the solution image. Quadratic penalty (QPs) functions are very popular due to their implementation simplicity. However, they produce images with non-uniform resolution characteristics. We have previously proposed new median based priors and evaluated their resolution characteristics. Here, we have compared noise characteristics of quadratic and median based priors. Our results show that median based priors produce images with lower standard deviation than filtered back-projection method and better-recovered resolution than quadratic priors with a small improvement in the contrast recovery coefficients (CRC).