Expectation maximization (EM; Lange and Carson 1984) is based on a statistical description of the imaging process. It considers the noise distribution in the image and tries to find a solution that generates the observed result with maximum probability.