Detecting COVID-19 Infection using Chest X-Ray Images through Transfer Learning

BACKGROUND: Coronavirus disease (COVID-19) spread around the world, causing a global crisis. On March 20th, 2011, the World Health Organization declared the virus to be pandemic. More than 200 nations have been impacted by the pandemic, and as of 10 October 2020, there are currently more than 37 million confirmed cases and much more than 1 million fatalities. The infection with COVID-19 was caused by the SARS-CoV-2 virus. As a result of this acute illness, the human respiratory system deteriorated. COVID-19 patients might experience signs resembling pneumonia and perhaps other breathing problems disorders during the first five to eleven days following the illness. Because of this, persons suffering from conventional pneumonia may have their COVID-19 undiagnosed. Due to this, computer-aided diagnostic (CAD) systems might be recognized a helpful tool for medical practitioners who desire to employ diagnostic scanning technology to aid in the identification of pneumonia and Covid-19. Despite the fact that chest X - ray images diagnostics is relatively rapid, COVID-19 disease identification should be carried out explicitly by skilled specialists, this requires expertise and is a time-consuming procedure. Nevertheless, there are several many more radiologists than patients being watched. As a result, an existing problem driven by artificial intelligence (AI) is expected to assist practitioners in screening COVID-19 patients more quickly and reliably. Without such a mechanism, sick people are unlikely to be quickly detected, separated, and addressed [16]. METHOD: Doctors can get a precise pre-diagnosis through the assistance of some deep learning (DL) approaches. In spite of the limited number of labelled samples that are currently accessible, this study suggested a specific and precise method for identifying healthy people, COVID-19, and pneumonia using CXR pictures. An image classification model family is called EfficientNet. The first article to discuss it was EfficientNet: Rethinking_Model_Scaling for Convolutional Neural Networks (CNN). EfficientNet-B0, EfficientNet-B4, EfficientNet-WideSE-B0, and EfficientNet-WideSE-B4 models can all be trained using the given scripts.[17] EfficientNet-B4 is used to train and predict accurate results utilizing a transfer learning-based methodology. The EfficientNet-B4 model has a few more layers added in order to get the proposed design. RESULTS: With 5675 X-ray images, the suggested approach achieved a significant degree of training precision of 98%. (1583 - healthy, 577 - COVID-19, and 3515 - pneumonia). Furthermore, it evaluates Covid-19, Pneumonia, and Normal pictures as the testing efficiency of 98%, 95%, and 98%. CONCLUSION: Early diagnosis of COVID-19 is necessary for taking measures because of the high prevalence of coronavirus infection. The proposed method of analysis can successfully identify Covid-19 and pneumonic infections in early CXR pictures. The outcomes show that, especially considered to the preceding methodologies, the proposed effort produced the most favorable results.