Study of the convergence properties of the updating coefficients in the EM algorithm for PET

The authors have performed a series of simulation studies of the expectation maximization (EM) algorithm, using the Hoffman Brain Phantom. They investigated the problem of image deterioration after the iterative procedure has been carried out for an excessive number of iterations and they have studied the convergence characteristics of the updating coefficients in this algorithm. These coefficients can provide a normalized to unity projection of the reconstructed image vectors and valuable information (added noise, location of pixels etc.) can be extracted from their statistical properties. A stopping criterion can also be extracted based on the results of this study.< >