Diagnosis of the Stage of COVID-19 Disease from CT Scan Images of the Lung by Using a Swin Transformer

COVID-19 is an infectious disease that emerged in China in 2019 and became a global pandemic. To contain the virus, governments adopted the strategy of testing, tracing, and isolating the infected people. This strategy required two steps: first, to detect the COVID-19 cases, and second, to assess the disease severity (which was vital for treatment). Various diagnostic methods were applied, but the most reliable one was analyzing the lung computed tomography (CT) images of the suspected cases. However, some radiologists lacked the experience with the new patterns and the high volume of patients increased the error rate. Therefore, an automated diagnosis system was needed. In this paper, we propose a system that uses the Swin transformer, a vision transformer model, to identify the disease severity of COVID-19. The system can distinguish between normal and abnormal lungs and categorize the disease severity to assist the doctor in treating the patient. We fine-tuned the pre-trained Swin Transformer network on our private dataset. We achieved 97% accuracy on our test dataset. Our dataset consisted of five classes (lung CT scan), including one normal class and four classes with different levels of involvement: (1) early stage, (2) progressive stage, (3) peak stage, and (4) absorption stage.