Research And Application Of Chest Radiography Classification Based On Generative Adversarial Networks

In today’s society,people’s living standards are constantly improved,and people are paying more and more attention to their own health.When people do physical examination,the chest X-ray image analysis results require us to wait for a long time,and it is hoped that the image processing speed will be further improved on the existing basis.Moreover,in medical images,chest X-rays occupy an important position,and the number thereof often generates a large backlog in medical institutions,and there is an urgent need to use automation to speed up the processing of chest X-ray images.The artificial intelligence system is used in high-precision chest X-ray image processing,which greatly reduces the workload of hospital radiologists,allowing medical staff to treat more patients in the same time.It is better to classify chest X-ray images using a classifier trained by the generative adversarial networks than the traditional method.For chest X-ray image classification research,there is currently no way to achieve 100% accuracy.To this end,this paper proposes to use a new model based on the generative adversarial networks to replace the deep convolutional neural network(DCNN)for the classification research.This paper constructs a domain-adaptive generative adversarial networks(DM-GAN)model for chest X-ray image classification,and produces a classifier with excellent performance through training model.At the same time,in the network layer of the progressive growth generator and discriminator,a series of low-to-high resolution images are used for training.During the training,we will add new convolution layers to make the details of the model better and better.The effect of the training comes better,the generator can generate excellent chest X-ray images,and the classification accuracy of the discriminator can be improved.During the model’s training,there is an imbalance in the distribution of chest X-ray image data sets,which indicates that the scale of healthy medical image data is very large,while the scale of pneumothorax,pulmonary edema and other image data is very small.This paper solves this problem by adding synthetic medical images to the data set.The DM-GAN model is trained with a combination of real and artificial image training data sets and a new training method.It not only balances the data set and improves the scarcity of some kinds of medical image data,but also improves the quality of the generated image through more rapid and stable training,which makes the discriminator easier to distinguish.The main contents of this paper are as follows:First,I introduce the research prospects of this topic and techniques related to chest X-ray classification.Describes what is the structure and characteristics of generative networks and adversarial networks.The application process of the deep convolution generative adversarial network(DCGAN)based on the generative adversarial network in the chest X-ray images is explained,and the advantages in dealing with the problem are explained.Secondly,this paper proposes the generative adversarial networks based on domain adaptive for chest X-ray medical image classification(DM-GAN).And we introduce the model structure of DM-GAN,describe the model structure of generative model and discriminative model in details,and explain the training process.Finally,a comparative test was conducted.The effect of the classifier produced by training DM-GAN is compared to the effect of the classifier generated by training DCGAN in chest X-ray image classification.By the comparison,it directly shows that DM-GAN is of great significance in chest X-ray applications,reducing the workload of medical profession and reducing people’s waiting time,especially for patients,further reducing the harm of the disease to them.