Metastasis Detection of Breast Cancer using Ensemble Deep Learning

Abstract

Breast cancer is one of the diseases which mainly affects women and is responsible for most of the deaths in Brazil, followed by skin and lung cancer. Among the consequences of occurrence, there are genetic predisposition, sedentarism, and late menopause, for example. The metastatic stage of this illness has a low survival rate because the disease spreads from the breast to other parts of the body, and the patients need the diagnosis as fast as possible to start the treatment. Moreover, state-of-art works claim that pathologists can reach 0.72 AUC in analyzing an exam composed of thousands of histopathologic images of lymph node sections. In this context, this work presents an Ensemble Convolutional Neural Network with Transfer Learning, called U-net VGG19, for detection using the PatchCamelyon dataset. Results indicate that the proposal reached an AUC of 0.9565 and a loss of 0.2869, reaching better results than state-of-the-art CNNs such as VGG16, VGG19, MobileNetV3Large, ConcatNet, and a custom-made CNN.