Quality improvement of reconstruction images in a PET system with nonuniform sensitivity due to its geometrical arrangement

As we develop new asymmetric positron emission tomography (PET) detector geometries, the nonuniformity of detector sensitivity is increased. In addition, when we exceed the number of iterations for image updating in the conventional reconstruction method, maximum likelihood-expectation maximization (ML-EM), the noise characteristics of reconstruction images become worse. We therefore usually terminate iteration at an early step. However, if we employ ML-EM in a PET system with nonuniform sensitivity, artifacts appear in the reconstruction images in the initial stage of iteration. In this report, we present a method for suppressing artifacts in the early steps of iteration in ML-EM by controlling the difference in the convergence speed referred to the log-likelihood and discuss the possibility of improving PET image quality based on computer simulations. (author)