Breast Cancer Detection using digital histopathology images and pre-trained deep learning models

Abstract

Cancer is a disease that can start anywhere in the body. It begins when infected cells grow out of control, outpacing normal cells. Breast cancer is the most common type in women around the world. Most of them are carcinomas, these originate in the cells that cover the organs and tissues of the body. The procedures used to detect the disease are diagnostic approaches, and some are invasive. Using digital tools, it is possible to develop or implement assisted diagnostic systems to streamline the process and allow greater reliability of the analyzes. The present study is carried out with digital histopathology images. In this study three scenarios were evaluated, starting from a classical scheme, then we include the use of pre-trained deep models and finally a deep model based on a convolutional neural network. The performance for each approach was evaluated by calculating three diagnostic measures such as precision, sensitivity and specificity. It is observed that the pre-trained models provide highly disciminative information despite having been trained for a completely different task. In general, deep models allow to significantly improve the specificity of the system when compared with the classical approach.