Mathematical morphology (MM) features for classification of cancerous masses in mammograms

Abstract

One of the important attributes of cancerous masses is their malignancy as it suggests a rapid growth of the cancer and possibility of metastasis. Malignancy, which denotes a special pathology of the tissue, is closely related to the existence of quasi-linear structures (spicules) emanating from the central mass. Hence, the tasks of malignancy and spicularity assessment are very often treated jointly. We propose a novel set of features enriching already existing pool of features for classification of masses. Our features are based on simple MM operations, pixel counting, and some basic algebra. To be more specific, given a contour of a cancerous mass we compute a sequence of dilations, and then count the number of pixels on the inner and the outer contour of each dilation. The contour pixel numbers are plotted against the size of the disk-shaped structuring element. The MM features are calculated from the plot via simple algebraic operations. The crucial point is that all the features are zero iff the input contour is circular. This distinctive property forms a basis for successful classification with the A z values higher than for the features existing in the literature. The additional advantage of our approach is the simplicity of the proposed features. In contrast to many features found in the literature, no sophisticated algorithms are employed, so reimplementation of the features should be easy for anyone interested. © 2008 Springer-Verlag Berlin Heidelberg.