Feature Contours Identification and Mammary Pathologies Detection in Mammography Images

Abstract

This paper proposes an algorithm for identifying regions of interest by inverse contrast enhancement that provides a representation of the different contours (mass, pectoral muscle, breast outer contour and Mcs). The second contribution is the detection of masses and micro-calcifications by manual thresholding and prewitt detector. This algorithm was tested using mammography images of different densities from multiple databases in a health clinic and images taken from the internet (40 images in total). The results are highly accurate, allowing better identification of the different contours, and better detection of breast pathologies (mass and micro-calcification). Finally, the identification of breast contours was performed using inverse contrast enhancement, which is the input of a manual thresholding algorithm specially designed for this purpose. After segmentation by manual thresholding, morphological opening, morphological dilatation and Prewitt contour detection we have a demarcation of the breast masses and micro-calcification. The results obtained show the robustness of the proposed enhancement and manual thresholding method. In order to evaluate the efficiency of our pathology detector, we performed a qualitative evaluation with a rate of 99% for the identification of the different contours and 98% for the detection of breast pathologies. A radiologist from the health clinic evaluated the results and considered them acceptable for the CAD.