Este repositório contém uma investigação sobre o uso de algoritmos de Aprendizado Profundo em tarefas de diagnóstico radiológico. Em particular, o foco é a classificação de distúrbios pulmonares em radiografias de tórax, usando técnicas Redes Neurais Convolucionais e Processamento de Imagens Digitais.
Radiology is a medical specialty that relies on extracting information from images to diagnose diseases and guide the treatment of patients. This task can be challenging as it requires a great deal of specialized knowledge and practical experience. Fortunately, Deep Learning offers a powerful way to automate image analysis and provide doctors with valuable information to help with diagnosis.
This repository aims to explore the use of Convolutional Neural Network models to classify lung disorders on chest X-rays. In particular, the disorders studied are Opacity, Atelectasis, Pneumothorax and Pleural Effusion. For this, Digital Image Processing techniques, reuse of Convolutional Neural Network architectures and Classification Committees are used.
The models constructed present an Accuracy of 87.77%, Accuracy of 91.96%, Sensitivity of 90.24% and F1-Score of 91.07% in the multilabel classification of chest X-rays. In addition, an interpretability analysis of the models is performed, using Activation Map algorithms to extract useful visual information and compare the results with examinations reported by radiologists.