文章詳目資料

Annals of Nuclear Medicine and Molecular Imaging

  • 加入收藏
  • 下載文章
篇名 列運算最大相似度演算法在Inveon MicroPET影像重建研究
卷期 21:3
並列篇名 A Study of Row-Action Reconstruction Algorithm for Small Animal PET Imaging Using Inveon MicroPET
作者 許若瑋許靖涵蕭穎聰林昆儒林名釗
頁次 131-139
關鍵字 序列子集均值與最佳化演算法最大相似度與均值最大化演算法列運算演算法小動物的正子斷層掃描儀OSEMMLEMRAMLASmall animal PET
出刊日期 200809

中文摘要

近年來列矩陣運算重建法在正子斷層掃描(PET)之應用發展迅速,例如序列子集均值與最佳化演算法(OSEM)由於大幅提升收斂速度,已廣泛使用於臨床PET。然而對小動物正子斷層掃描(microPET),由於光子數限制造成較大的影像雜訊,因此如何選擇快速且穩定的影像重建方法,仍是一項很重要的議題。本研究針對Inveon microPET探討列運算最大相似度演算法(RAMLA)之效能和可行性,也與最大相似度與均值最大化演算法(MLEM)及序列子集均值與最佳化演算法比較。本研究取Inveon microPET掃描小鼠的正弦圖,以MLEM、OSEM及RAMLA重建,MLEM疊代次數?32次;OSEM序列子集階層?2、4、8、20、40,疊代次數?8、20次;RAMLA因定鬆弛係數?0.5、1、1.5,序列子集階層4,疊代次數?8次。鬆弛係數在列運算演算法中扮演一個重要的控制者的角色,調控每次疊代中修正的幅度。從結果中可發現OSEM重建的影像與其他兩種演算法相比,兼具著高速率與穩定性,在相同的資料運算量及耗時下,RAMLA更適用於microPET。

英文摘要

Recent advances of the row-action algorithms, like ordered subsets expectation maximization (OSEM) algorithm, have greatly facilitated statistical PET image reconstruction into clinical practice. However, for small PET studies, the photon counts are usually limited and inevitably lead to a noisier image reconstruction. Therefore, how to choose a fast and stable statistical reconstruction method still remains an important issue. In this study, we implement and investigate the performance of a more advanced row-action maximum likelihood algorithm (RAMLA) and compare its performance to OSEM and MLEM on an Inveon microPET for small animals. We used a clinical rat PET study acquired and reconstructed the rat sinogram using MLEM, OSEM, RAMLA. For an objective comparison, the iteration numbers of these methods are: MLEM with 32iterations, OSEM with 4-subset 8 iterations and 2, 4, 8, 16, 20, 40-subset and 20 iterations, RAMLA with 4-subset and 8 iteration and relaxation parameter 0.5 and 1.5, respectively. Relaxation parameter in RAMLA serves as a controller which adjusts the amount of image update at each iteration. From the experimental results, the RAMLA produces image results with faster convergence rate than MLEM and maintains stable convergence compared to OSEM. Under the same amount of computation time, RAMLA would be an appropriate and versatile reconstruction method for small animal PET.

相關文獻