A deep learning model for CXR-based COVID-19 detection

Coronavirus disease (COVID-19) pandemic has mobilized the world since the beginning of 2020, and is still globally disturbing our lives. Its diagnosis being usually unclear for doctors, patients are considered as positive cases by default, entailing a costly treatment. It was established that CXR images, i.e., those obtained by chest X-rays, are efficient in COVID-19 diagnosis. Therefore, automating the CXR image processing for early disease diagnosis is crucial for radiologists. The main contribution of the paper is the development of a novel deep convolutional neural network (CNN) model, used to discriminate between positive and negative cases. In addition, the flip, rotation, zoom and shift operations were used to augment the dataset size. Finally, the model was trained and tested on COVIDx CXR-2 dataset. The respective percentages were: 96.75 for accuracy, 96.99 for precision, 96.50 for recall, 96.98 for sensitivity, 96.52 for specificity and 96.74 for F1-score; which compare very well with state-of-the-art models.