Breast Abnormalities' Classification Using Convolutional Neural Network

Abstract

Deep learning (DP) holds great promise in many areas, especially in the medical field. Abnormalities in the breast threaten patients' lives, so it is crucial to go through the correct diagnosis. Thus, the participation of a convolutional neural network (CNN) in image analysis and classification could support a proper diagnosis. A set of mammograms are collected in the CBIS-DDSM dataset to train different CNN architectures and various models to classify the labeled images into mass and calcification as a first step and benign or malignant as a second step. The first task is accomplished using two different models, the first one is a CNN and the second includes the VGG16, both achieved good results on the validation with accuracies of 88% and 90% respectively. Regarding the second task, it is performed using CNN only. The accuracy does not exceed 66% due to the limitation in the number of mammograms.