Patient imaging systems, such as PET imaging systems, for example, may suffer from the introduction of artificially introduced noise. This noise is, typically, introduced during iterations of reconstruction algorithms, such as the least-squares algorithms, which attempts to recreate a 2D or a 3D image from raw acquisition information. The noise appears as “hot-spots” in the reconstructed image. Approaches to address these artefacts use filtering approaches. Typically, a least-squares reconstruction is supplemented with a penalty term, an approach known as “Relative Difference Penalty”. The penalty parameter causes the reconstruction algorithm to filter more or less strongly at certain regions of the reconstruction. The present application proposes an approach which supplements the penalty term with continuous probability information about the likelihood of an edge being present in a portion of an image.