The prevalence of breast cancer in the modern world has motivated the development of new tools to assist radiologists in their quest to detect malignancy as early as possible. Following the successful introduction of the screening programmes, science must provide effective clinical methods to detect cancer and improve life expectancy. Considerable research has been undertaken to this end, but the results still lack the robustness necessary for routine clinical applications. Mammographic images are difficult to interpret even by radiologists and this makes their task error prone. This paper presents a new approach to filtering breast images, which highlights the structures of anatomical interest. A method to detect calcifications has been explored. The approach is based on an edge preserving filtering with anisotropic diffusion. The algorithm makes use of the advantages offered by the hint images, a normalised physical-based representation of the breast. The results are promising with excellent true positive rates in both detection of isolated coarse calcifications and microcalcifications with a very low number of false positives per image.