Research on Classification of ConvNeXt Chest X-ray Images Based on Attention Mechanism

Abstract

The global novel coronavirus pandemic has become the norm, and lung disease is one of the most important diseases at presentation In order to improve the reading efficiency and the accuracy of disease classification, this paper proposes a transfer learning model based on attention mechanism: CS ConvNeXt In the data preprocessing stage, the generalization ability of the model is improved by data augmentation, and in the model learning stage, the ConvNeXt network with attention mechanism is used to study and migrate to the proposed dataset COVIDx, and the model is tested by the COVIDx dataset in the testing stage Experiments show that the CS ConvNeXt transfer learning method can improve the accuracy of CXR image classification, and the average accuracy of Chest X-ray images for three categories reaches 98.18%, of which the accuracy for COVID-19 reaches 99.52%.