Channel Attention Residual Network for diagnosing Pneumonia

Lately, the rampant of Coronavirus (COVID-19) epidemic of global has drawn a lot of attention to pneumonia and other infectious diseases of the lungs. chest X-ray images are the most effective way to diagnose and detect pneumonia diseases. In this paper, a channel attention residual network model ECA-XNet is proposed for diagnosing pneumonia with chest X-ray images. Using deep residual networks instead of deep convolutional neural networks, which can effectively prevents problems such as network overfitting. The pre-trained ResNet model with weights and parameters are migrated to the model. Then, the channel attention module ECA is introduced into the residual structure, which enhances the learning of local features by the residual network. The proposed model has been validated on the Chest X-Ray Images dataset, and the 50-layer ECA-XNet network based on 50 layers showed the optimal performance with better accuracy, sensitivity, specificity and average F1-score than other baseline models, The average accuracy of ECA-XNet(50) is 96.30%, which is more than 1% higher than ResNet (50). On the whole, A good classification ability is obtained in the pneumonia diagnosis.