Uso de técnicas de inteligência artificial, correlação de imagens térmicas e do conceito de impedância térmica visando a estimativa da localização e do tamanho de tumores mamários

Abstract

Breast cancer has the highest incidence and mortality in the female population worldwide. Early and accurate detection of breast cancer is a critical part of the strategy to reduce the mortality associated with this prevalent disease. Mammography is the most used technique for the early detection of breast cancer. However, it has several limitations. Several images are needed, and there is a strong dependence on the operator. There is still a firm reliance on a qualified physician to identify tumors on X-ray images. In addition, lifetime exposure to X-rays can also induce tumors. In this project, we propose the development of techniques based on applying the thermal impedance method and on detecting inclusions based on the use of correlations of surface temperatures of the skin of the breast to detect the origin of the heat source (abnormal metabolism of cancer). The main idea of this work is to use the artificial intelligence method, deep learning, a technique related to the recognition of specific features in images to detect sources of metabolic heat. In addition to the proposed approach not being invasive or causing pain to the patient, it must allow access to people with disabilities or low mobility, have low cost and use national technologies. The low sensitivity to tiny and deep tumors, typically found in the analysis of surface temperatures using thermal imaging, is circumvented by utilizing the concept of thermal impedance and artificial intelligence techniques, such as deep learning. It has been shown a brief theoretical foundation on convolutional neural networks, optimizers, activation functions, and the hyperparameters that must be adjusted in the neural network. The development of the thermal model and the creation of the database from its solution are described, as well as the choice of parameters detectable in thermographic images, deep learning libraries, network training using convolutional neural networks, and the analysis of the hyperparameters used in the model for the convergence of the forward and inverse problems. We also discuss the need for prior knowledge of tumor characteristics that can be estimated via convolutional neural networks and the influence of tumor size variation by comparing the temperature and thermal impedance signals. The numerical results of estimating the location of tumors in numerical simulations using thermographic images obtained from simulated data in a Cartesian model and an anatomical breast model are shown. The development of a technique that makes it possible to identify the location and size of breast tumors from information on temperature and surface heat flux is an unprecedented and promising tool in the early detection of breast cancer.