The Study Of Three-dimensional CT Image Reconstruction Using Algebraic Methods And Its Hardware Acceleration

It is known that the algebraic-based methods are important image reconstruction techniques for X-ray transmission tomography. They have their own advantages for these situations where it is not possible to measure a large number of projections or the projections are not uniformly distributed. However, the complicated structure and large computational complexity obstruct the algebraic-based methods'application in practice, especially for the case of cone-beam reconstruction. Thereby whether for academic research or for practical application, it is valuable to study how to improve the computational efficiency of the reconstruction methods and find a befitting way to apply the reconstruction methods fast.In this paper, two CT image reconstruction techniques, which are able to improve computational efficiency, are proposed. And an accelerating frame for the algebraic-based methods based on commercial video card is designed. The proposed accelerating method can reduce the computational time for cone-beam reconstruction greatly. So it carves out a feasible way for practical applications of the algebraic-based methods.This paper proposes two reconstruction methods: adaptive simultaneous algebraic reconstruction technique (ASART) and support region based CT image reconstruction technique. The ASART takes advantage of the multilevel scheme(MLS) projection access order,simultaneous update and adaptive weight contribution, and has fast convergence. Results reconstructed from both synthetic and practical CT data show that the reconstructed images obtained by ASART after one or two iterations are better than these obtained by MLS-SART after ten or even more iterations. The support region based CT image reconstruction technique reconstructs the object using a constraint of the object support region, which is obtained by a optical measurement equipment or a low-energy X-ray transmission equipment, so as to reduce the number of projections used. In comparison with the traditional reconstruction method without the support region constraint, simulations show that the support region based method can obtain the reconstructions with same image quality by using only one-third to a half of the projections which are used by the traditional reconstruction method.In the hardware acceleration aspect, the algebraic-based methods are transplanted to general purpose computing platform based on commercial video card. Characteristics of GPU's stream processing and video card's high bandwidth are used to speedup the reconstruction process. Sphere sampling in forward projection based on 3D texture and algebraic-based reconstruction technique based on Flat3D texture are implemented. It solves the problem of reconstruction algorithms'hardware compatibility, which makes it possible to perform cone-beam 3D image reconstruction on ordinary PC. Experiment results show that it takes only 1 second for the hardware acceleration method to reconstruct an object of 128×128×128 in one-iteration, and this is about 1/92 of that reconstructed by the corresponding software reconstruction method.