Segmentation Of Spiculated Masses In Mammographic Images

Breast cancer is among the leading causes of cancer deaths for women. In 2012, 522 000 deaths have been recorded worldwide, representing a 14% increase compared to 2008. Generally, the benign masses class is associated with the presence of circular or oval shapes, while spiculated masses are more likely to belong to the malignant masses class. Thus, spicule is a leading discriminant factor in the classification of various masses. Its extraction is a complex task because of their low contrast, variable widths and the overlapping of blood vessels, fibers and ducts. With the increase of images obtained during the screening, mammography interpretation by radiologists is becoming more difficult, time-consuming, and leads sometimes the increase the ratio of false positives due to tissue superimposition. Hence, to help radiologists improve detection and diagnosis accuracy the design of computer aided detection systems (CADe) known as a great leap forward in recent years due to their ability to provide an objective and reproductible second opinion. The CADe are structured in three steps: segmentation of the region of interest which contains the mass and description of the segmented mass. Based on these steps, the proposed method for automatic breast mass detection can be described as follows: the segmentation is based on MRF using the Pickard random field (PRF) which is much faster, more robust and nearly unsupervised compare to most of MRF-based methods, which require complex and time-consuming computations [1, 2]. The description step presents the main contribution of this paper, since, to the best of our knowledge, we present the first attempt to extract the spicules with the mixture of a Markovian framework and an a contrario model.