Abstract:
Along with a health crisis, COVID-19 has also led the world towards an economical barrier. So far the virus has effected approximately 400 Millions causing 5 Million deaths and is expanding everyday. There is an urge to stop the exponential growth of the contagious disease, only possible through an early diagnosis of the disease. Currently, several testing techniques are being used to diagnose COVID-19, among them Polymerase Chain Reaction (PCR) is a gold standard globally. However, due to it’s processing time, cost and less sensitivity towards COVID19, physicians suggest to correlate the results with radiological tests preferably Chest X-Ray (CXR) imaging since it consumes less time and is more sensitive towards COVID-19. To overcome the pandemic many research groups have been working on the solution. Several Computer Aided Diagnostic (CAD) systems have been proposed by the researchers however, they lack robustness and stability towards blind datasets. Moreover, majority of the CAD systems provide binary classification between healthy and COVID-19, various lung abnormalities resembles COVID-19 in terms of their structural appearance and can be falsely classified as COVID-19. In this paper, we have proposed a deep model using EfficinetNet-B0 as a baseline model. Our proposed model has been trained on the largest available CXR dataset of COVID19 comprising CXR images of normal, Viral Pneumonia, Lung Opacity and COVID-19 effected lungs and yielded an accuracy of 99.46%. Proposed model has been blind tested on four publicly available datasets achieving highest accuracy of 99.96%. Furthermore, the model is transfer learned and fine tuned on another publicly available CXR dataset and evaluated to be 85.26% accurate for 20 epochs.
Page(s):
1-1
DOI:
DOI not available
Published:
Journal: IEEE International Conference on Digital Futures and Transformative Technologies (ICoDT2) May 24-26, 2022 (Book of Abstracts), Volume: 1, Issue: 1, Year: 2022