Deep Learning-Based COVID-19 Detection from Chest X-ray Images: A Comparative Study

Abstract

The COVID-19 pandemic, caused by the SARS-CoV-2 virus, has rapidly spread across the globe, leading to a significant number of illnesses and fatalities. Effective containment of the virus relies on the timely and accurate identification of infected individuals. While methods like RT-PCR assays are considered the gold standard for COVID-19 diagnosis due to their accuracy, they can be limited in their use due to cost and availability issues, particularly in resource-constrained regions. To address this challenge, our study presents a set of deep learning techniques for predicting COVID-19 detection using chest X-ray images. Chest X-ray imaging has emerged as a valuable and cost-effective diagnostic tool for managing COVID-19 because it is non-invasive and widely accessible. However, interpreting chest X-rays for COVID-19 detection can be complex, as the radiographic features of COVID-19 pneumonia can be subtle and may overlap with those of other respiratory illnesses. In this research, we evaluated the performance of various deep learning models, including VGG16, VGG19, DenseNet121, and Resnet50, to determine their ability to differentiate between cases of coronavirus pneumonia and non-COVID-19 pneumonia. Our dataset comprised 4,649 chest X-ray images, with 1,123 of them depicting COVID-19 cases and 3,526 representing pneumonia cases. We used performance metrics and confusion matrices to assess the models' performance. Our study's results showed that DenseNet121 outperformed the other models, achieving an impressive accuracy rate of 99.44%.