Research Of Fast 3D Image Reconstruction Methods Based On The Programmable Graphics Hardware

With the increasing of 3D CT's scan speed, image reconstruction speed becomes the efficiency bottleneck in practical cases instead of the scan speed. Especially in some cases that real-time reconstruction is required, the image reconstruction efficiency does not meet the requirements, which will limit the application of 3D CT. Recently the performance of the graphics hardware is promoted greatly, and shows great possibilities to be as a general purpose processing units, which makes the research of general purpose computations based on GPU (Graphics Processing Units) become a hot research topic. Aiming at improving the image reconstruction speed, we put forward two fast GPU-based reconstruction methods in this paper. Taking into account of utilizing all the computing resources of PC, we also researched the parallel 3D image reconstruction algorithm. The main research contents and achievements are articulated as follows:1. Design of an object-oriented software development toolkit (SDK) of the programmable graphics hardware. Using the object-oriented techniques and the design patterns, we designed an object-oriented SDK through encapsulating the Cg function library. Under the help of the SDK, researchers will promote their development efficiency and can devote more time into their research work.2. Research of fast image reconstruction methods based on GPU. In this paper, we first researched the fast image reconstruction method based on the projective texture mapping, through mapping the computing patterns to projective texture mapping. Then we optimized this method by transferring all the computations from CPU to GPU, which promotes the reconstruction speed greatly. At the same time, we put forward another fast image reconstruction method based on the geometric parameters lookup-table. It implemented multi-channel reconstruction and fully utilized GPU's computing capacity.3. Research of the parallel image reconstruction method. Aiming at solving the problem of huge-data reconstruction, we discussed two divide and conquer strategies. Based on these two strategies, we researched the CPU-GPU hybrid fast image reconstruction method and the parallel reconstruction method based on the PC cluster, which utilizes all the computing resources of PC and it will be an effective way to reconstruct huge-size images.4. Design and implementation of the 3D image reconstruction platform. To improve the research work's development efficiency using high level language for CT algorithms, we show the design of a half-open platform for CT algorithms, which will free researchers from the heavy coding works.It is shown that the speed of the GPU-based fast 3D image reconstruction methods that are introduced in this dissertation, promotes greatly compared with the original FDK method and it can satisfy the practical requirements.