Explainable Deep Learning for Pulmonary Disease and Coronavirus COVID-19 Detection from X-rays.

• We show how deep learning can be applied for COVID-19 detection from chest X-rays; • The proposed method is aimed to mark as first step a chest X-ray as related to a healthy patient or to a patient with pulmonary diseases, the second step is aimed to discriminate between generic pulmonary disease and COVID-19. The last step is aimed to detect the interesting area in the chest X-ray (to provide explainability); • We propose an explainable method aimed to automatically detect the areas of interest in the chest X-ray, symptomatic of the COVID-19 disease. • We obtain an accuracy of 0.99 in COVID-19 detection by evaluating 6,113 chest x-rays, with a time window required for the detection approximately equal to 2.5 seconds. Background and Objective : Coronavirus disease (COVID-19) is an infectious disease caused by a new virus never identified before in humans. This virus causes respiratory disease (for instance, flu) with symptoms such as cough, fever and, in severe cases, pneumonia. The test to detect the presence of this virus in humans is performed on sputum or blood samples and the outcome is generally available within a few hours or, at most, days. Analysing biomedical imaging the patient shows signs of pneumonia. In this paper, with the aim of providing a fully automatic and faster diagnosis, we propose the adoption of deep learning for COVID-19 detection from X-rays. Method : In particular, we propose an approach composed by three phases: the first one to detect if in a chest X-ray there is the presence of a pneumonia. The second one to discern between COVID-19 and pneumonia. The last step is aimed to localise the areas in the X-ray symptomatic of the COVID-19 presence. Results and Conclusion : Experimental analysis on 6,523 chest X-rays belonging to different institutions demonstrated the effectiveness of the proposed approach, with an average time for COVID-19 detection of approximately 2.5 seconds and an average accuracy equal to 0.97. [ABSTRACT FROM AUTHOR]