The Development Of Low-dose CT Image Reconstruction System Based On Generating Antagonistic Network

Although X-ray Computerized Tomography(CT)is able to provide abundant information of human anatomy,the complete CT scan of the X-ray may cause ionizing radiation damage to patients.Reducing the amount of X-ray in the direction of scanning angle and detector can effectively decrease the radiation dose;however,it will result in the partial absence of the projection data,and finally leading to the appearance of serious artifacts on the CT reconstruction image,which will affect the clinician's diagnosis and treatment to the location of the lesion.Therefore,how to reconstruct high quality CT images with incomplete CT projection data is of great significance.The purpose of this article is to develop a low dose CT image reconstruction algorithm and system based on generating adversarial network.In this article,a series of antagonistic networks are constructed to estimate the missing projection data in scanning angle direction and detector direction,so as to complete the missing projection data and improve the image quality of CT reconstruction.The main work of this article is as follows:(1)The basic principles of CT imaging system and adversarial network are introduced,and the research status of low-dose CT reconstruction algorithm and adversarial network is introduced in detail.(2)A low-dose CT image reconstruction algorithm based on adversarial network is proposed.The algorithm constructs a generation network which is similar to an automatic encoder in the training phase.The generated network uses an encoder on basis of convolutional neural network to learn the structural information of incomplete projection data,and then missing part of incomplete projection data is estimated by the decoder.Then the probability of whether the missing partial projection data is true or not is estimated by the discriminant network of the same structure.After obtaining the complete projection data,the CT image is reconstructed by the filter back-projection reconstruction algorithm.Experiments show that low-dose CT reconstruction algorithm based on adversarial network can effectively eliminate artifacts in CT images and improve the quality of reconstructed image under low-dose conditions.(3)A low-dose CT reconstruction algorithm based on joint loss function is proposed.Low-dose CT reconstruction algorithm based on generating adversarial network is not able to capture the high frequency information of real projection data,therefore,this paper improves the algorithm and constructs a joint loss function composed of reconstruction loss and adversarial loss.The projection data estimated by the adversarial network on basis of the joint loss function is not only consistent with the real projection data generally,but also consistent with the real projection data at the single pixel point.Experiments show that the low-dose CT reconstruction algorithm based on joint loss function can obtain better reconstruction results,especially in the case of lower radiation dose.(4)A low dose CT image reconstruction system based on adversarial network is constructed,which integrates all the functions of the algorithm proposed in this article.In this article,we use the adversarial network to improve the image quality of CT reconstruction in the absence of partial projection data,and achieve the purpose of reducing the injury of ionizing radiation and obtaining CT images that meet the needs of clinical diagnosis.