Pneumonia Recognition in Chest X-rays through Convolutional Neural Networks (CNN)

Abstract

Pneumonia is a major threat to respiratory health and can be caused by bacterial or viral illnesses. Untreated pneumonia can be fatal, underscoring the significance of early diagnosis. The goal of this work is to automate the process of discriminating between pneumonia caused by bacteria and viruses utilizing digital X-ray pictures. The text commences with a comprehensive summary of the progress made in enhancing the precision of pneumonia diagnosis. It then proceeds to elucidate the approach employed by the writers. The four distinct deep convolutional neural networks (CNNs) employed for transfer learning are SqueezeNet, AlexNet, ResNet18, and DenseNet201. A set of 5247 pictures, including chest X-rays displaying bacterial, viral, and normal states, were preprocessed and used transfer learning to train the classification task. The authors of the study have delineated three methods for classifying pneumonia: distinguishing between normal and viral, differentiating between bacterial and viral, or identifying a combination of all three. 98% of the photos correctly classified as both viral and bacterial pneumonia, 95% for both normal and pneumonia instances, and 93.3% as all three forms of pneumonia. Its performance substantially outperforms the previously known accuracies. Therefore, the suggested study may aid radiologists in diagnosing pneumonia more quickly and expedite airport screenings for individuals with the illness.