Automatic region of interest segmentation in thermal images of the breast using deep networks

This repository contains the code developed for my undergraduate thesis in Computer Science at the Federal University of Uberlândia.

Link to UFU Repository: https://repositorio.ufu.br/handle/123456789/39556

Abstract

Breast cancer is the most incident type of cancer in women in Brazil, with a forecast of over 73 thousand new cases expected for the year 2023. Early detection of this type of cancer significantly increases the chances of successful treatment. In this context, infrared thermography stands out as a non-invasive technique that can be applied even in younger women for breast cancer detection. The analysis of thermographic breast images for breast cancer detection is a topic addressed in the literature, and in this sense, the segmentation of the region of interest in these images aims to remove some information that may be considered as noise for the automatic detection of the disease. This work proposes a method for segmenting thermographic breast images using the DeepLabV3+ Convolutional Neural Network, achieving average accuracies, Intersection over Union, and precision of 98.69%, 97.18%, and 98.48%, respectively.

Key words

• Image segmentation
• Breast cancer
• Infrared thermography
• Region of interest
• Convolutional neural network

Reference

PINTO, Tiago da Silva e Souza. Segmentação automática de região de interesse em imagens termográficas da mama utilizando redes profundas. 2023. 36 f. Trabalho de Conclusão de Curso (Graduação em Ciência da Computação) - Universidade Federal de Uberlândia, Uberlândia, 2023.