One-class Classification for Identifying COVID-19 in X-Ray Images

Coronaviruses constitute an extensive family of viruses that can be severely harmful to both animals and humans. The newest virus of this family, SARS-CoV-2, and its associated disease in humans, COVID-19, have become a worldwide problem that requires bringing together different strategies to deal with it. The affectations of COVID-19 largely vary among individuals, ranging from a lack of symptoms to death. One of the fingerprints of COVID-19 is the damage caused to the respiratory system, which is often diagnosed based on a chest X-ray. In this work, we present an approach for classifying chest radiographs to identify the presence of COVID-19. Three different one-class based classifiers were implemented, and different image pre-processing techniques were applied to the radiographs to identify the combination of pre-processing/classifier that leads to the best results. For experimental purposes, we make use two datasets: one containing images from patients with COVID-19, and the second one with chest X-ray images corresponding to patients diagnosed with various acute respiratory conditions as well as healthy patients. The obtained results validate the feasibility of using the proposed methodology as an aid in the diagnosis of COVID-19.