Automatic classification of breast tumors using circularly approximated contour

In this paper a simple, novel and automatic method for classification of breast malignancy in digital mammogram images is proposed. It is based on the similarity/differences between the contour extracted from the mammogram image and its nearest circular approximation. The nearest circular approximation of extracted contour is obtained by considering the centroid of tumor as its centre and arithmetic mean of maximum and minimum radial distances of contour points from the centroid, as its radius. The similarity between the fitted circle and tumor contour is measured in terms of variance of radial distance of contour from the centroid, as a feature for classification. The simulation results show that for a set of 150 tumor contours, the proposed method gives 96.67% accuracy. The performance obtained in terms of the receiver operating characteristic (ROC) parameters like accuracy (Ac), sensitivity (Se), specificity (Sp), and positive (PPV) and negative predictive values (NPV) are 96.67%, 0.9873, 0.9437, 0.9512 and 0.9853 respectively.