Diagnosis of Breast Cancer on Full Mammography Using the BRINT Texture Descriptor and Convolutional Neural Networks

Breast cancer is one of the leading causes of mortality among women. Early detection of this disease is crucial and mammographic analysis is a fundamental tool for its diagnosis. Classification of mammograms is a complex task due to the need to identify the presence of tumors (benign or malignant) as well as the absence of tumors. In the literature it is common to group the classes into categories such as abnormal (malignant or benign) and normal, also classified into malignant and non-malignant (benign and normal), and there are even works that exclude the normal class. The paper proposes a methodology that combines the BRINT descriptor with convolutional neural network (CNN) models for the classification of full mammograms into benign, malignant, and normal. It has been demonstrated that the methodology proposed in this work has been shown to have 96.7% on average of accuracy and f1 score for the following databases: MIAS, Mini-DDSM, INbreast, and KAU-BCMD.