Detecting COVID-19 Lung Damages from Chest X-Rays Images by AI-based Algorithm

Coronavirus Disease 2019, is an infectious respiratory disease caused by the virus called SARS-CoV-2 which in November 2020 recorded more than 60 million cases and approximately 1.45 million deaths worldwide. The most widely used diagnostic test is the Reverse transcriptase-polymerase chain reaction (RT-PCR), accompanied by chest X-Rays (CXR) designated as the main diagnostic tool in high-risk areas and for symptomatic subjects awaiting response from the RT-PCR test for which up to 24 hours are required. Although CXR is a quick and valid diagnostic test, its discrimination against viral pneumonia is the main clinical challenge, due to the similar lung dam- age that is reported in alveolar tissue in the early stages of COVID-19. In this study, has been proposed an automated algorithm for detecting COVID-19 lung damages from CXR images using ResNet convolutional neural networks: ResNet-18 and ResNet-34. In particular the supervised learning technique has been followed in order to learn the network in the COVID-19 and viral pneumonia recognition using the CXR images database present in the Kaggle CXR image repository. After the training both the networks have been tested on a test dataset providing the following results: accuracy of 98.33%, precision of 100.00%, sensitivity of 96.67%, specificity of 100% and F1-score of 0.98 for the ResNet-18. The results of the ResNet-34 have been the following ones: accuracy of 95.00%, precision of 100.00%, sensitivity of 90.00%, specificity of 100% and F1-score of 0.95 for the ResNet-34. In conclusion the ResNet-18 provided better results of the ResNet-34 resulting the best candidate for the COVID-19 lung damage detection and discrimination from viral pneumonia being an optimum tool for overcoming the main clinical challenge.