A Lightweight AMResNet Architecture with an Attention Mechanism for Diagnosing COVID-19

Abstract

Aims: COVID-19 has become a worldwide epidemic disease and a new challenge for all mankind. The potential advantages of chest X-ray images on COVID-19 were discovered. We proposed a lightweight and effective Convolution Neural Network framework based on chest X-ray images for the diagnosis of COVID-19, named AMResNet. Background: COVID-19 has become a worldwide epidemic disease and a new challenge for all mankind. The potential advantages of chest X-ray images on COVID-19 were discovered. Objective: A lightweight and effective Convolution Neural Network framework based on chest X-ray images for the diagnosis of COVID-19. Methods: By introducing the channel attention mechanism and image spatial information attention mechanism, a better level can be achieved without increasing the number of model parameters. Results: In the collected data sets, we achieved an average accuracy rate of more than 92%, and the sensitivity and specificity of specific disease categories were also above 90%. Conclusion: The convolution neural network framework can be used as a novel method for artificial intelligence to diagnose COVID-19 or other diseases based on medical images.