Research And Acceleration Of Exact Reconstruction Algorithm In CT Image Reconstruction

With the development of the medical equipment industry, CT machines has become animportant tool for doctors to diagnose. The technology of Computerized Tomographyincludes data acquisition,correction, image reconstruction and many other algorithms. Imagereconstruction algorithm is to reconstruct the CT images for diagnosis from acquisition data.This part of CT technology is one of the most critical steps because the performance ofreconstruction algorithm will affect the CT image directly. The traditional filteredback-projection algorithm which based on the Fourier slice theorem is wildly used incommercial CT machines because of it’s excellent reconstruction ability in2-dimension and3-dimension image reconstruction.The noise of CT image can be reduced by improving FBP algorithm, however, thisalgorithm can not solve the problem of region of interest reconstruction by improving it’sperformance. And there is another difficult problem in fast image reconstruction which cannot be solved by any improvement algorithm of FBP. This article is just focus on these twoproblems which are improving of CT image reconstruction quality and region of interestreconstruction.Spiral beam BPF algorithm is able to reconstruct ROI problem with high calculatedquantity. Based on the existing research of project, this paper studies the improvement ofparallel beam reconstruction algorithm. The BPF image reconstruction algorithm in spiralbeam is applied to parallel beam scanning mode which reduce the computation and improvethe quality of image at the same time. BPF reconstruction algorithm of parallel beam is ableto achieve ROI reconstruction because of the steps of BPF algorithm. Two parameters ofimage assessment field are calculated to evaluate reconstruction performance withShepp-Logan mode by the improved algorithm. The results indicate that the improvedalgorithm is able to reconstruct the image accurately, and the result is closer to the humanvisual system.The fan-beam is the mainstream of modern CT machine scanning mode, therefore, thisarticle try to study two different kinds of optimization idea of fan-beam reconstructionalgorithm.Idea one is to improve the filter function according to the traditional FBP algorithm.The original filter function is replace by derivative and Hilbert transform which is called DHBalgorithm and Shepp-Logan image is reconstructed through Matlab simulation by theimprovement.The result shows that the improvement can achieve the goal of increasing peak signal to noise ratio by using the same image assessment algorithms mentioned above,Another idea is to reconstruct the image by using PI lines instead of traditional FBP algorithm.So the parallel-beam BPF algorithm is introduced into the fan-beam reconstruction, theassessment results show that the improved algorithm of fan-beam BPF algorithm based on PIlines is closer to human visual system compared with traditional FBP.The ultimate purpose of this work is applied to the actual dental CT. But the purpose ofrapid reconstruction can not be done just rely on the improvement to the algorithm withoutsome technologies of other fields. Therefore, the content of research in this part is to studyparallel computing of Graphics Processor Unit in the field of computer at last. ComputeUnified Device Architecture which is called CUDA is applied to the two kinds of improvedfan-beam image reconstruction algorithm. This technology improved parallel computingpower of algorithm without affect the precision of reconstruction algorithm. Actual data isapplied to show that the technology and improvement of reconstruction algorithms in thispaper are able to achieve the purpose of ROI reconstruction and improve the reconstructionspeed11.42times.