This research focuses on a state-of-the-art deep learning model for sorting chest X-ray (CXR) images among four classes: Normal, Pneumonia, COVID-19, and Other Lung Diseases. The MIA team souped up VGG19 and MobileNet designs with Multi-Head Attention tricks to get better at extracting features and zeroing in on the areas of interest that are crucial for spotting diseases. They trained their models on a data set of 15,000 tagged images; they cleaned it up through standardization and other manipulation in order to make the models perform better over all. All models we worked shows the much better results of Mobile Net it shows the 98.9% accuracy 0.97 precision, and 0.96 recall by adding the attention mechanism it made to improve and precise by diagnosis of COVID -19. In addition to the MobileNet, too, got up to speed more quickly and required less computing power compared with the others, making it a better fit for on-the-spot use. This work shows how attention-empowered lightweight models could help simplify the way doctors diagnose problems, take some of the pressure off radiologists and extend better care to places that don’t have abundant resources. The researchers’ next steps will be to further refine the model, expand the dataset, and then deploy it in a real world application to assist with automated diagnosis support.