Covid-19 Detection from Chest X-Ray Images Using Deep CNN Architectures with Transfer Learning

Abstract

The novel coronavirus COVID-19 first appeared in China at the end of 2019 and was subsequently classified as a world pandemic. At the time of writing, the number of affected persons is 52,331,462 persons and the number of deaths is 1,287,966 deaths. The most used screening methods of COVID-19 is Reverse-Transcription Polymerase Chain Reaction (RT-PCR) test. The number of RT-PCR test kits available is limited because of the increasing number of cases. Some people with COVID-19 have difficulty breathing and their lungs are damaged. Consequently, radiologists utilized Chest X-Ray images to detect the damage caused the COVID-19 into the lungs. However, manual detection takes an important time and depends on the radiologist's expertise. Therefore, it is important to implement automatic detection methods to solve this problem. Due to the limitation of data sets containing COVID-19 images and the small number of training data, transfer learning based on Convolutional Neural Networks (CNN) can be a good combination to solve this problem. In this work, we propose two pre-trained CNNs architectures AlexNet and Residual Network (ResNet-50) to detect COVID-19. The two presented architectures are trained to detect COVID-19, normal and pneumonia from Chest X-Ray images using a 10-Fold cross validation method. Our proposed model outperforms the existing methods and yielded a mean classification accuracy of 96,74% with AlexNet and 99,2% with ResNet-50. In the future work, we will increase the number of COVID-19, Normal and Pneumonia images in the datasets to outperform the performance metrics.