ABSTRACT
The entire world suffers from a health crisis in the midst of a global COVID-19 pandemic. Researchers are trying to address their concerns by providing remedies to save lives and halt the pandemic outbreak. As early detection and diagnosis of the cases slows down the spread of this disease among the people, there is a need of an automatic detection of COVID-19. Creating a workable algorithm to detect and assess COVID-19 severity using the chest X-ray (CXR) obliges a vast amount of carefully chosen COVID-19 datasets, which are hard to obtain. This circumstance is appropriate for the Swin Transformer architecture because it enables the self-attention mechanism to utilize a significant amount of unlabeled information through structural modeling. The backbone network’s inherent attributes are employed as inputs in Transformer architecture for COVID-19 identification and to estimate severity. To assess our model’s generalization ability, we use a variety of test datasets. The testing results demonstrate that the proposed approach obtained the best results in both detection and diagnosis tasks and yields an average precision value of 0.93 as a result.
