Research On Reconstruction Algorithm And Visualization Of Cone-beam CT Based On GPU

Computed Tomography(CT)is a technology that can reveal the internal structure of an imaged object through X ray scanning.It is an important branch in the medical image field,and is widely used in industry,prospection,space and other fields.Cone beam CT(Cone-beam CT,CBCT)is one of the CT technologies.Because of its fast acquisition speed,high spatial resolution and high radiographic efficiency,it has become a hot topic in recent years.However,the amount of data obtained by CBCT scanning is huge,the complexity of reconstruction operation is relatively high,and the reconstruction time is relatively long.In recent years,with the rapid development of Graphics Processing Unit(GPU),especially the CUDA(Computer Unified Device Architecture)programming architecture developed in 2006,it promotes the wide applications of parallel computing based on GPU,and provides a new path for the parallel acceleration of the CBCT reconstruction algorithm.This paper mainly aims at the GPU parallel acceleration of CBCT reconstruction algorithm,and initially realizes the 3D visualization of CBCT reconstruction results.The specific work is as follows:1.The cone beam FDK(Feldkamp-Davis-Kress)algorithm and the feasibility of parallel computing were analyzed.First,the physical and mathematical basis of CT imaging is summarized.Secondly,starting from the two-dimensional fan beam back projection reconstruction algorithm,the FDK algorithm of 3D cone beam CT is illustrated in detail.Finally,the feasibility of parallel computing for FDK algorithm was analyzed.So as to lay a theoretical foundation for the next step of parallel accelerated computing.2.The GPU parallel acceleration of FDK algorithm was realized based on the CUDA framework.The FDK algorithm in the Descartes coordinate system is improved by using the characteristics of the trigonometric function periodicity,and the parallel acceleration of the algorithm is designed and implemented by using CUDA technology.In the weighted calculation,each GPU thread calculates a pixel of the CT projection,and realizes the parallel computation of all pixels.In the filtering process,the image data is rearranged to reduce the calculation times of Fourier transform,improve the computing speed,and improve the data transmission speed by using constant memory.In the process of back-projection,using the independent characteristics of the rebuilt data and the periodicity of the trigonometric function,the calculation times of the trigonometric function are greatly reduced.Twelve projections are calculated in parallel using GPU to accelerate the back-projection procedure.The experimental results show that,with the reconstructed volume of 512 ? 512 ? 512,the acceleration ratio is more than 310 times compared to that calculated using CPU without reducing the image quality.The experimental results show that,comparing with the traditional CPU-based FDK reconstruction algorithm,the proposed GPU-based CBCT reconstruction algorithm has improved the reconstruction speed by more than 310 times while guaranteeing the quality of the reconstructed image.3.Based on Marching Cubes and visualization package VTK,a simple 3D visualization software is designed by QT.The software has four display windows,which can display 3D volume and cross sections including transvers,sagittal and coronal planes.It also supports the translation,rotation and scaling of the 3D volume data and 2D graphics.