COVID-GATNet: A Deep Learning Framework for Screening of COVID-19 from Chest X-Ray Images

摘要

In the preceding several months, the outbreak of COVID-19 has become one of the global severe public health issues. It is critical to screen for COVID-infected or suspected patients in time. Given that the method of nucleic acid test strip detection is laborious and time-consuming, the Chest Xray images screening method has become the object of this study because of its robust availability and easy accessibility. This study introduces a new neural network model, COVID-GATNet, to assist radiologists in automatically diagnosing CXR images, thus improving the detection speed of suspected infected people. The study integrates three CXR data sets published on the internet and Kaggle competition, including CXR images of healthy, other types of pneumonia, and COVID-19 positive patients. Because there is less open-source data for COVID-19 positive CXR images than the other two types of data, this research dilated CODIV-19 positive CXR images by scaling, rotating, adjusting brightness and other increase methods of image data. Then, the expanded COVID-19 positive CXR data set is divided into a training set and test set at a ratio of 8:2 to complete the experiment and evaluation of COVID-GATNet. Research by combining Densenet Convolutional Neural Network (DenseNet) and Graph Attention Network (GAT), COVID-GATNet has a direct connection between any two network layers in each dense block and uses the attention mechanism, which significantly reduces the parameters of the model and improves the classification performance. Experimental results show that the accuracy of COVID-GATNet is best to achieve 94.1%, and the F-1 score also with 95.2%. Compared with COVID-Net, COVID-Net accuracy is 93.3%, and the F-1score is 94.8%. The experimental results and diagrams show that the proposed model performs well and can effectively assist clinicians in screening and detecting patients' state, helping to classify pulmonary diseases and observe the pulmonary status of patients.