A COVID-19 CXR Image Recognition Method Based on Deep Transfer Learning

Although the immediate threat of the coronavirus disease 2019 (COVID-19) has been significantly mitigated through widespread vaccination efforts and public health measures, ongoing observation and preparedness remain crucial in managing potential future outbreaks and ensuring global health security. Our research focuses on using a transfer learning-based convolutional neural network (CNN) to distinguish and classify chest X-ray images (CXR) of patients with COVID-19, pneumonia, and relatively normal lungs. The proposed transfer learning model, named NewNet, integrates features from CXR images and deep learning model, AlexNet [1]. Training of the transfer learning CNN model was conducted using two datasets: the Kaggle CXR dataset [2] and the dataset collected by Joseph et al. [3]. The experimental results indicate that the NewNet model achieves an average accuracy of 96.82% for classifying COVID-19 CXR images, 92.94% for pneumonia, and 96.32% for normal cases. Thus, through the transfer learning technique, the proposed NewNet deep learning network demonstrates improved accuracy in diagnosing COVID-19.