Detection of COVID-19 from radiographic images using convolutional neural networks: A bibliographical review

Abstract

The crisis generated on the planet by COVID-19 (SARS-CoV-2) caused a devastating effect worldwide, and for this reason, an effective detection of the possible contagion of infected patients was needed. In this sense, the present work gathers information from diagnostic tools that use Deep Learning (DL) in medical images to detect COVID-19. It is a descriptive observational study. In addition, the purpose of this study is to analyze and compare how DL applied to radiographic images optimizes resources and management of results in an objective and timely manner, showing a favorable cooperation between the health, institutional and technological sectors. In such a way that Convolutional Neural Networks (CNN) in their different algorithms are the chosen architecture in the biomedical area for the diagnosis of diseases applied to the analysis of radiographic images, which purpose is to help the medical service to lighten the attention of patients with an early detection of symptoms and risk factors of the COVID-19 virus, due to the number of symptomatic and asymptomatic patients. The results of this Systematic Literature Review show the degree of accuracy of the use of neural algorithms when evaluating medical images. Therefore, it is concluded that CNNs have generated very useful results to issue a timely diagnosis when validating positive cases of COVID-19, but it is evident that in most of the reviewed works, an evaluation protocol that overestimates the results has been applied.