應用圖形處理器技術於電腦斷層掃描重建之研究

Abstract

電腦斷層掃描技術可以廣泛應用在許多領域，例如：醫學上的放射檢查、機場海關的檢查等等；在電腦斷層影像的重建技術中，以濾波反投影以及疊代式兩種演算方法最為著名。濾波反投影演算法的優點是計算時間快速，但是重建出來的影像會因為計算上的誤差而品質較差；疊代式演算法源自於數值分析，需要耗費大量的計算時間，但是重建出高品質的影像為其優勢。近年來，愈來愈多研究利用圖形處理器平行計算的特性，幫助我們處理大量資料的計算。因此，本篇論文中利用NVIDIA所開發的整合技術CUDA，將疊代式重建法中可以平行處理的部份實作於圖形處理器中執行，在計算時間上比在CPU中執行快了數十倍之多；並且更應用三角函數和角公式的特性，讓演算法在圖形處理器中更有效率地執行。實驗結果中以各種模擬與實際資料，比較計算效能上的改善，以及兩種不同演算法所重建出來的影像品質。最後，本篇論文利用QT的開發平台，將各種CT重建方法整合成可以跨平台的使用者視窗介面。

Computerized Tomography can be applied in many fields, such as medical radiology and Customs check in airport. Both Filter Back Projection and Iterative Image Reconstruction is well-known algorithm in the scope of CT image reconstruction method. The advantage of Filter Back Projection is very fast, but the image reconstructed usually suffers from noise problem because of the inaccuracy in computing process. The Iterative Reconstruction originated in numerical methods, which can reconstruct high quality image but require heavy computation power. In recent years, more and more researches take advantage of parallel computing property in GPGPU, GPGPU help us to handle large number computation power problem. Therefore, we utilized the integrated technique named CUDA which is developed by NVIDIA Corp in this paper. We implemented the parallel computing of Iterative Reconstruction Method in GPGPU, the expense time is better than it is executed in CPU by a factor of tens. Furthermore, Sum and difference formulas in Trigonometric functions are applied in order to let algorithm execute more efficiently in GPGPU. As the experiment result, we compared the improvement of performance between CPU and GPGPU and the quality of image between Filter Back Projection and Iterative Reconstruction Method. Finally, we utilize QT creator to integrate CT reconstruction methods in cross-platform user interface.