New mass description in mammographies

Abstract

In this article, we present a new mass description dedicated to differentiate between different mass shapes in mammography. This discrimination aims to reach a better mammography classification rate to be used by radiologists as a second opinion to make the final decision about the malignancy probability of radiographic breast images. Therefore, we used a geometrical feature which is perimeter measurement (P) and 3 morphological features which focus on mass borders by discriminating circumscribed from spiculated shapes. These features are: contour derivative variation (CDV), skeleton end points (SEP) and we propose a new one noted Spiculation (SPICUL). Their performance were evaluated one by one before collecting them for mammography classification into the 4 BIRADS categories. For classification, we used support vector machine (SVM) with Gaussian kernel as classifier for its higher performance. The accuracy of our model with contour features for classifying malignancies was 93% in the case of two class model (malignant and benign) and 85.7% in the 4 class model (BIRADS I,II,III and IV).