PneumoNet: Automated Detection of Pneumonia using Deep Neural Networks from Chest X-Ray Images

Abstract

Pneumonia is a dangerous disease that causes severe inflammation of the air sacs in the lungs. It is one of the infectious diseases with high morbidity and mortality in all age groups worldwide. Chest X-ray (CXR) is a diagnostic and imaging modality widely used in diagnosing pneumonia due to its low dose of ionizing radiation, low cost, and easy accessibility. Many deep learning methods have been proposed in various medical applications to assist clinicians in detecting and diagnosing pneumonia from CXR images. We have proposed a novel PneumoNet using a convolutional neural network (CNN) to detect pneumonia using CXR images accurately. Transformer-based deep learning methods, which have yielded high performance in natural language processing (NLP) problems, have recently attracted the attention of researchers. In this work, we have compared our results obtained using the CNN model with transformer-based architectures. These transformer architectures are vision transformer (ViT), gated multilayer perceptron (gMLP), MLP-mixer, and FNet. In this study, we have used the healthy and pneumonia CXR images from public and private databases to develop the model. Our developed PneumoNet model has yielded the highest accuracy of 96.50% and 94.29% for private and public databases, respectively, in detecting pneumonia accurately from healthy subjects.