Feasibility of Breast Cancer Detection Through a Convolutional Neural Network in Mammographs

Abstract

In the Iraq female samples, the malignant neoplasm type with the highest mortality rate is breast cancer. When the disease is detected early, the success rate is higher, resulting in improved prognosis and, consequently, cure. The present study aims to analyse the viability of a system capable of detecting breast cancer through convolutional neural networks, classifying a mammogram into five classes: non-cancer; benign calcification; malignant calcification; benign mass; and evil mass. Processing was carried out on the database containing 55,890 images, in which the data was converted from records structure to image format, which is necessary when using the neural network. After this stage, the images were classified into the five categories mentioned to enable tests to be carried out to verify the accuracy of the machine learning algorithm in identifying and classifying breast cancer. Using a small partition of 10% of images from the total database to verify the initial results presented in this work, it was possible to obtain 44% of global accuracy, highlighting the ability to expedite the early and rapid detection of breast cancer using artificial intelligence.