Redes neurais densamente conectadas para detecção de câncer de mama em imagens histopatológicas

Abstract

Even with all the advances in medical technologies, cancer still is one of the biggest causes of death on earth. Women breast cancer is second cancer with more incidence if diagnosed early can be cured. One procedure to the diagnosis of cancer is the biopsy, which produces images that must be analyzed through a microscope by a pathologist. With computational vision and deep learning advances, it’s possible to have computer help doctors for a better rate of correct cancer diagnosis, with the experience of the doctors and the computational power of actual machines, the life quality of the people which suffer from cancer can be improved, doing diagnosis with more precision. The DenseNet is one architecture of the artificial neural network that emerged from these advances presented by Huang et al. (2017), aiming a better performance, has connection patterns never seen before between the layers, handles very well with the vanish gradient problem and requires fewer parameters. The following paper presents key concepts, techniques, and implementations to achieve this objective in relation to medicine, it will be used deep learning, neural network and image processing techniques that showed the viability of applying deep learning in medicine to recognize biopsy images of the breast region that may contain cancer Were made binary and multi-class classification in this work.