A New Architecture For Diagnosing Pulmonary Thorax Diseases (Covid-19, Pneumonology, Normal) Using Deep Learning Technology

Abstract

Medical imaging is an efficient field of research because it makes it possible to diagnose and detect diseases and relieve patients by presenting a remedy or treatment to follow. Pulmonary pathology represents a principal challenge for doctors to diagnose and examine the type of disease. The use of medical x-ray images has made it possible to advance research in terms of medical findings by specialists. Despite this, doctors sometimes misdiagnose the patient, and we suggest employing artificial intelligence to significantly enhance the patient’s medical diagnosis and help identify lung ailments more accurately. In this project, we plan to develop a computer system based on deep learning with a coherent approach based on the application of standard convolutional neural networks (CNN) with architectures appropriate and specific to this problem (VGG16, MobileNetV1, ResNet50, InceptionV3, DenseNet121, and DenseNet169). We also propose the construction of an architectural model called RafikNet. The tests will be realized on a corpus of images with three types of lung disease (COVID-19, pneumology, and normals), and we will evaluate our approach on parameters such as accuracy, precision, recall, f1-score, support, and loss.