The Research Of CT Reconstruction And Visualization Based On GPU

CT(Computed Tomography)is a imaging modality that can obtain internal information of measured objects by the X-ray scanning,which has been widely used in the fields of aerospace,medical,machinery,and industry etc.With high resolution and shorter scanning time,Cone-Beam CT has attracted much attention and been the focus of research.Due to the large-scale data and high computation complexity,how to speed up the cone-beam CT reconstruction on PCs is research hotspot.In recent decades,with the development of the Graphics Processing Unit(GPU),and CUDA programming architecture,GPU programming has been more easy and convenient for developers.Meanwhile,the hardware platform and software in 3D visualization of cone-beam CT are expensive for users.In this paper,an acceleration strategy based on FDK algorithm and a visualization software were developed.The main works are outlined as follows:1)An acceleration strategy based on FDK algorithm:In FDK reconstruction process,three progress of weighting,filtering and backprojection were speed up by using CUDA programming,respectively.In weighting progress,the data of each angle was processed by using parallel computing.In filtering progress,the CUFFT library was used,and the weighted data was rearranged to reduce invalid computing and storage space;In back projection progress,the FDK algorithm was speed up by using the symmetries(including vertical and rotational symmetry),and the four texture channels of CUDA architecture.On the premise of ensuring the quality of the reconstructed image,the speed of the GPU can be increased by more than 100-200 times than that of the CPU,and the acceleration ratio was higher when the experimental data is increased.2)A visualization software:Based on VTK,CTK and QT,a visualization software was developed on Visual Studio 2013.The software mainly used the ray-casting algorithm of volume rendering,which was divided into two parts: display window and operation window.The display window was mainly composed of four parts.You can view two-dimensional data of different sections of any layer and perform scaling and rotation operations on the three-dimensional volume data.The operation window supports dynamic adjustment of ambient light,diffuse reflection,and highlights.For every voxel,you can click on the mouse to display its location,gray value,etc.