CT Image Reconstruction With Missing Projection Data Based On The Framework Of Adversarial Learning

X-ray Computed Tomography(CT)is a common clinical diagnostic tool that can provide clear images of human anatomy.Although limiting the scan view of the CT equipment can effectively reduce the ionizing radiation received by the patient,it will result in partial missing of projection data at certain scanning angles,and ultimately lead to a substantial decrease in the quality of the reconstructed CT image.Therefore,how to reconstruct high-quality CT images that meet the needs of clinical diagnosis when the CT projection data is incomplete has important social value and theoretical research value.For the purpose of improving CT reconstruction with missing projection data,this paper proposes a series of limited-view CT image reconstruction algorithms under the framework of adversarial learning.Firstly,a deep learning reconstruction algorithm for limited-view CT images based on perceptual loss is proposed.It takes part of the missing projection data at the scanning view as input,and uses a generative adversarial network structure similar to an autoencoder in the training phase to generate missing CT projections.The encoder is used to learn and extract the feature information of the incomplete projection data,and then the decoder predicts the missing part of the projection data from the extracted features.Then,a discriminator network with a structure similar to that of the encoder is used to show whether the predicted missing data can be regarded as real data.In order to improve the prediction accuracy,this paper also adds perceptual loss to the loss function through the pre-trained VGG network.The trained generation network can predict the missing part of the newly collected incomplete projection data,and finally use the filtered back projection algorithm to reconstruct the CT image from the completed projection data.The experimental results show that the method in this paper can effectively improve the quality of CT image reconstruction when the projection data is missing.Secondly,this paper also puts forwards a CT image reconstruction algorithm based on a U-shaped adversarial automatic encoding network because the U-net network structure can effectively integrate the multi-scale features in medical images.This model,aiming to combine the advantages of U-net with that of generative adversarial networks uses its unique U-shaped structure and layer-jumping connection mechanism to fuse the feature maps of CT projection data at different scales,and then combines the adversarial ideas of generative adversarial networks to improve CT image reconstruction.Finally,starting that the prediction result should be consistent with the real data,this paper also proposes a CT image reconstruction algorithm on the basis of the consistency of the projection data.By combining the characteristics of the sinusoidal image,the Helgason-Ludwig consistency condition is added to the network,so that the predicted missing part of the projection data can be consistent with the known one,thereby effectively ensuring the quality of the CT reconstruction which meets the requirements of clinical diagnosis while reducing the X-ray dose when the projection data is incomplete.