A Revolutionary Approach to COVID 19 Detection Using a Novel Transformer Based Architecture

Abstract

The Covid19 pandemic continues to be a global crisis, resulting in tragic loss of millions of lives. This has spurred researchers to explore deep learn-ing (DL) techniques for Covid 19 diagnosis, aiming to assist medical pro-fessionals in the screening process offering valuable second opinions to clinicians. To that end, we introduce a novel DL architectural design for Covid 19 detection, which combines the strengths of vision transformers (ViTs) for capturing long range dependencies with Efficient Net's (EffNet) fine-grained classification capabilities. Built upon the EffNet‑B0 backbone, the ViT-based model with the ViT-B/16 configuration extracts global con-text and long-distance feature information from input images, yielding powerful feature representations. 94.78 % is the accuracy achieved with our proposed model, demonstrating its effectiveness following experimental verification. The efficacy of our model has been empirically substantiated in comparison to state-of-the-art (SOTA) approaches. Despite its initial fo-cus on natural language processing (NLP), our substantial accuracy demon-strates that the ViT model exhibits promising performance and holds great potential for broader applications in computer vision (CV) tasks.