Deep Learning Applied to COVID-19 Detection in X-Ray Images

Introduction

In Wuhan 2019, the emerged the coronavirus disease (COVID-19) (Roosa et al., 2020)⁠ caused by the virus severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) (Stoecklin et al., 2020), which is the third highest pathogenic coronavirus detected in humans preceded by the severe acute respiratory syndrome coronavirus (SARS-CoV) and the middle east respiratory syndrome coronavirus (MERS-CoV). On March 11, 2020, the World Health Organization (WHO) declared the COVID-19 outbreak a pandemic (Zhang et al., 2020)⁠. An infected person spreads the SARS-CoV-2 virus when coughs or sneezes primarily through droplets of saliva or discharge from the nose (World Health Organization, 2020). Infection can occur within 14 days of exposure but in most cases in no more than 4 to 5 days (Guan et al., 2020)⁠. The economical and health consequences of the rapid spread of the virus worldwide forced the adoption of extraordinary social confinement measures to halt its dissemination and to prevent the collapse of the health systems (Martínez Chamorro et al., 2021). It is estimated that between 30–40% of the COVID-19 patients are asymptomatic (Oran & Topol, 2020)⁠, others can experience broad symptoms severities. Most of the cases only presented mild or moderate symptoms (Yang et al., 2020)⁠, the most common are fever higher than 38°, cough, myalgia, smell and/or taste abnormalities, and headache (Martínez Chamorro et al., 2021). However, a study reports that 15.7% of patients admitted into the hospital developed severe illness (Guan et al., 2020)⁠, including severe pneumonia, pulmonary edema, acute respiratory distress syndrome, multiple organ failure, among others (Mo et al., 2020)⁠. The early identification and diagnosis are important to mitigate the COVID-19 outbreak (Guan et al., 2020), which as of July 16, 2021, the number of worldwide cases is 188,655,968, including 4,067,517 deaths (World Health Organization, 2021).

A report from the WHO solidarity consultum from February 2021 evaluates multiple drugs and shows that the mortality, initiation of ventilation, and hospitalization duration were not definitely reduced by any trial drug. Until now, no specific drug has been found against COVID-19 (WHO Solidarity Trial Consortium, 2021)⁠. Reverse transcription-polymerase chain reaction (RT-PCR) is among the most efficient and reliable methods for the detection of SARS-CoV-2, it is performed using a sample of nasopharyngeal or respiratory secretions. This method has a high specificity and sensitivity that can range from 60–70% to 95–97%, it can vary depending of the time elapsed since exposure to the virus (Martínez Chamorro et al., 2021)⁠. However, it takes hours to provide a result and needs specialized equipment and personal. The rapid antigen detection (RAD) tests are used to overcome these limitations since it is fast, easy to perform and interpret (Rathore & Ghosh, 2020)⁠.