The Coronavirus Disease 2019 (COVID-19) pandemic keeps on spreading across the globe and the expeditious spread of COVID-19 since its outbreak has impelled many nations healthcare frameworks & economy to the edge of breakdown. Therefore, to suppress the spread of the disease and minimize the ongoing expenditure on the healthcare system, accurate identification & isolation of COVID-19 positive individuals for treatment is vital. Reverse Transcription Polymerase Chain Reaction (RT-PCR) has been the golden standard for diagnosing COVID-19 due to its accuracy & authenticity but is expensive and requires trained professionals, laboratory, and RT-PCR kit for COVID-19 detection & analysis which is time-consuming & arduous. Medical images such as chest X-rays (CXR) are crucial to confirming a COVID-19 diagnosis, as they provide accurate laboratory test as visual evidence to physicians and radiologists while being readily accessible in most healthcare systems. Various research on COVID-19 detection and treatment based on emerging technology is still ongoing. In the last few years, researchers have exhibited utilization of Deep Learning (DL) methods like Convolutional Neural Network (CNN) on chest X-ray (CXR) for COVID-19 detection which speeds up the COVID-19 diagnostic process, but CNN methods fail to capture the global context due to their inherent image-specific inductive bias.

Therefore, a need to develop advanced Deep Learning model using Transformers for COVID-19 detection arises. The project is based on detecting & classifying between normal, pneumonia, and COVID-19 patients using Chest Radiography (X-ray) images, using different Transformers utilizing attention mechanism where we will use publicly available COVID-19 datasets (like COVIDx CXR-2, VGG-16, MobileNet-v2, COVID-Net, EfficientNet-B7 etc.) for training & fine-tune the transformers model for the multiclass classification problem (COVID-19, Pneumonia and Normal cases) as per required. We will include results on the basis of performance metrics (Accuracy, Precision (Positive Prediction Value), Recall (Sensitivity), F1 score, Specificity, MCC or Negative Prediction Value (NPV)) and try to show that the features learned by our Transformer networks are explainable. Furthermore, we will compare the results with previously implemented state-of-the-art methods using Transformers for COVID-19 detection on different datasets.