ACRnet: Adaptive Cross-transfer Residual neural network for chest X-ray images discrimination
Abstract Background: Cardiothoracic diseases are a serious threat to human health and chest X-ray images have great reference value for the diagnosis and treatment. However, it is difficult for professional doctors to accurately diagnose cardiothoracic diseases through chest X-ray images sometimes and there will be different understanding based on human subjective differences, which will affect the judgment and treatment of diseases. Therefore, it is very necessary to develop a high-precision neural network to recognize chest X-ray images of cardiothoracic diseases.Methods: In this work, we cross-transfer the information extracted by the residual block and by the adaptive structure to different levels, which avoids the reduction of the adaptive function by residual structure and improves the recognition performance of the model. To evaluate the recognition ability of ACRnet, VGG16, InceptionV2, ResNet101 and CliqueNet are used for comparison. In addition, we use the deep convolution generative adversarial network (DCGAN) to expand the original dataset.Result: We find ACRnet has better recognition ability than other networks in identifying cardiomegaly, emphysema and normal. Besides, ACRnet's recognition ability has been greatly improved after data expansion. Conclusions: The experimental result indicates that emphysema and cardiomegaly can be effectively identified by ACRnet. Using DCGAN technology to expand the dataset can further improve the recognition ability of the model.