LUISA: A breast cancer Computer-aided diagnosis proposal on scanned mammography images. A breast cancer Computer-aided diagnosis proposal on scanned mammography images.

Abstract

A growing problem worldwide, breast cancer is considered a major cause of death in women. Digital mammography is the main method of early detection of this cancer, but its interpretation is difficult even for a professional. Machine learning techniques are used to facilitate this interpretation. Thus, the present work aims to propose a computer-aided detection system to collaborate with professionals in the diagnosis of breast cancer based on the analysis of scanned mammography images. Using the SURF key points algorithm and underwent a classification process with Convolutional Neural Networks (CNN) and Random Forest (RF) to generate a mass candidates dataset through the analysis of 1210 CBIS-DDSM base images. CNN presented better results reaching in training 0.06% loss and 0.97% accuracy. In the validation scenario, this proposal achieved 0.32% loss and 0.93% accuracy. For the RF model, the characteristics of the image extracted from the Hu-Moments descriptor getting an accuracy of 0.43% ± 0.005.