Low-dose Cone-beam CT Image Reconstruction System Based On 3D Adversarial Generative Network

Cone-beam Computed Tomography(CBCT)is an imaging technique that directly obtains threedimensional medical images.But Excessive doses of ionizing radiation increase the risk of cancer in the body.Although reducing the amount of cone beam CT projection data can effectively reduce the ionizing radiation to which the patient is exposed,it will cause a significant reduction in the quality of the reconstructed cone beam CT image.Therefore,how to reconstruct a high-quality cone-beam CT image that meets the needs of clinical diagnosis under the condition of low-dose CT(LDCT)has important research value.In order to improve the reconstructed image quality of finite-angle cone-beam CT,a finite-angle cone-beam CT reconstruction algorithm based on a three-dimensional adversarial generation network is proposed.The algorithm in this chapter can be divided into three stages: data preprocessing,training,and CT reconstruction.In the data preprocessing phase,the CBCT three-dimensional projection data with limited angles is cut into low-dimensional three-dimensional projection data matrix blocks as input data in the training phase.In the training phase,the purpose of generating a network is to generate real missing projection data,while discriminating the network is to distinguish the true and false projection data.During the reconstruction phase of the finite angle cone beam CT image,the new finite angle CBCT projection data is cut into low-dimensional three-dimensional projection data matrix blocks after data preprocessing,and this is used as input data for the trained three-dimensional generation network.The trained generation network can estimate the missing projection data part,and then stack the multiple three-dimensional projection data matrix blocks in order to obtain the completed CBCT projection data.Finally,cone beam CT images can be reconstructed using the completed projection data.The experimental results show that the method can effectively improve the quality of reconstructed CT images when the projection data is missing in the scan angle direction.Secondly,in order to improve the reconstruction quality of cone beam CT images of sparse projection data,this paper proposes a sparsely sampled cone beam CT reconstruction algorithm based on a three-dimensional adversarial generation network.The algorithm in this chapter can be divided into three stages: data preprocessing,training,and CT reconstruction.In the data preprocessing stage,the sparsely sampled CBCT three-dimensional projection data is processed into low-dimensional three-dimensional sparse projection data.In the training phase,the processed three-dimensional sparse projection data is used as input to train the generation network.In the CT reconstruction phase,the three-dimensional sparse projection data after the new data preprocessing is input to the trained generation network to estimate the missing part of the sparse projection data.Finally,the projection data estimated by the generated network and the sparse projection data of CBCT are combined to obtain the complete three-dimensional projection data of CBCT,and the cone beam CT image under sparse sampling is reconstructed by the FDK(Feldkamp Davis Kress)algorithm.The experimental results show that the method can still effectively improve the quality of reconstructed CT images when sparse sampling is performed in the scanning angle direction.