Application Research On View-driven CT Image Reconstruction Of Ultra-Sparse Projection

Computed Tomography(CT)is an imaging technology for obtaining cross sections of objects.According to the difference of absorptivity and transmittance of X-ray to human tissue or object's internal structure,firstly,using a highly sensitive detector to scan the object to be reconstructed to obtain projection data,and then putting the projection data into a computer,finally,the cross-sectional image of the object can be obtained after data processing by an electronic computer.CT imaging technology plays an important role in industrial detection and medical diagnosis,but there are also some problems in its practical application.In industry,due to the complex environment of the production site,the scanning method of sparse projection data can only be used for nondestructive testing.In medicine,reducing the view angles is usually adopted to shorten the detection time in order to reduce the harm of X ray to human body or to avoid the movement of the object during detection.Therefore,how to reconstruct the tomographic images from sparse projection data accurately enough to provide the reconstructed images suitable for industrial detection and imaging clinical diagnosis has become a hot issue in the field of CT imaging research.Based on existing literature and communicating with CT technicians,the projection data with no more than 20 view angles in fan-beam/cone-beam scanning in the range of[0,2?)are referred to as ultra-sparse projection data.CT image reconstruction algorithms can be divided into analytical algorithms and iterative algorithms.Analytical algorithms have the advantages of simple implementation,fast reconstruction speed and high imaging quality,but they are often used in image reconstruction of complete projection data,and the quality of reconstructed images obtained under the condition of sparse projection angles is poor.Iterative algorithms have advantages over analytical algorithms in sparse projection reconstruction.With the development of large-scale parallel computing technology and the reduction of computer hardware cost,iterative algorithms have become the focus of researchers in related fields and manufacturers of CT machines.In this paper,the problem of CT image reconstruction under ultra-sparse projection data is studied in the following aspects:(1)View-driven system model is proposed.Forward/back projection operation is the core part of CT image reconstruction,and the system model has an important influence on the numerical accuracy and image quality of iterative CT image reconstruction.There,combining the advantages of pixel-driven and ray-driven models,and based on the basic idea of the distance-driven model,we propose the view-driven system model.As well as combining CT iterative image reconstruction with compressed sensing theory,a two-dimensional fan-beam CT iterative image reconstruction algorithm is designed,which is named as 2D-CSVD(2D-Compressed Sensing View Driven CT image reconstruction)algorithm in this paper.(2)The 2D view-driven model is extended to 3D,and the 2D algorithm is extended to 3D CT image reconstruction of cone beam projection,which is referred to as 3D-CSVD algorithm.The CSVD algorithm consists of two steps:rough image reconstruction based on view-driven model and optimization calculation.(3)NVIDIA's CUDA architecture is used to carry out GPU heterogeneous parallel processing on the time-consuming 3D cone-beam CT image reconstruction program.The experimental results show that under the condition of ultra-sparse projection data,the proposed algorithm has certain academic research significance and engineering application value,which is embodied in:(1)high numerical accuracy.The reconstructed image under the ultra-sparse view angles can accurately reproduce the spatial structure and pixel distribution of the model image,and the minimum number of view angles that can be handled is 18.At the same time,the image quality indexes of the view-driven 2D-CSVD algorithm are significantly better than that of the ART-TV iterative algorithm and FBP analytic algorithm.(2)Low computational complexity.The view-driven forward/backward projection operation strategy in this paper can be summarized as:processing all pixels under one view angle in one iteration,which reduces a lot of unnecessary traversal operation and makes the computational complexity significantly lower than traditional iterative CT image algorithms.(3)Less memory overhead.In the traditional iterative CT image reconstruction,the scale of system matrix is usually huge,but the image reconstruction based on view-driven system model does not need to store the system matrix,which greatly reduces the memory overhead.Under the condition of ultra-sparse projection data,the CSVD algorithm proposed based on the compressed sensing theory and the view-driven model provides a practical technical way for researchers in related fields to carry out CT iterative image reconstruction under the condition of ultra-sparse projection data.