Mammographic Masses Descriptor for Breast Cancer Classification and Automatic Diagnosis
Mohammed El Amine Yermes, Mohammed Debakla, Khalifa Djemal · Revue d intelligence artificielle · 2024
An automatic breast cancer diagnosis is a challenging task because breast masses have a random appearance and vary in size and shape.A descriptor is an algorithm that quantifies elementary characteristics such as color, texture, contour, or shape.In digital mammography, numerous descriptors have been employed to differentiate between benign and malignant tumor patterns, but automatic diagnosis remains a difficult function.In this paper, we proposed a novel approach based on local features to describe masses in mammograms via Polygon Approximation Triangle-Area Representation (PATAR).As the degree of spiculation in masse defines their level of malignancy, the strength of our approach lies in its ability to isolate and measure spiculations in breast masses.PATAR is a robust image descriptor composed of two steps: polygon approximation and triangle-area representation.Firstly, we applied a polygon approximation to the masses to raise the most critical spiculations and lobulations.Then, by browsing the points of the polygon, calculate the triangle's area formed by the vertices of the polygon, ears, and mouths.The extracted characteristics describe the shape and show the severity of spiculations with high precision.Digital mammography CBIS-DDSM is used to evaluate the method with a Fuzzy C-Means classifier, Support Vector Machines (SVM), and Random Forest (RF).The Random Forest classifier achieved the best performance, reaching 97,94%.The proposed method provides a fully automated diagnosis with the best accuracy and invariance to scale and rotation.