Characterization and optimization of image quality as a function of reconstruction algorithms and parameter settings in a Siemens Inveon small-animal PET scanner using the NEMA NU 4-2008 standards
Characterization and optimization of image quality as a function of reconstruction algorithms and parameter settings in a Siemens Inveon small-animal PET scanner using the NEMA NU 4-2008 standards
复制标题
DOI：
10.1016/j.nima.2010.10.089
复制
发表时间：
2011-02-11
影响因子：
1.4
通讯作者：
Boerman, Otto C.
中科院分区：
文献类型：
--
作者：
Visser, Eric P.;Disselhorst, Jonathan A.;Boerman, Otto C.
Siemens Inveon小动物PET扫描仪提供的图像重建算法为滤波反投影（FBP）、三维重投影（3DRP）、二维或三维有序子集期望最大化（OSEM 2D/3D）和最大后验（MAP）重建。本研究旨在根据NEMA NU 4-2008标准定义的图像质量（IQ）优化重建参数设置。NEMA NU 4-2008图像质量体模用于确定图像噪声，表示为均匀体模区域的标准差百分比（%STDunif）。FOG填充棒的放射性恢复系数（RCrod），以及非放射性水和空气填充体模隔室（SORwat和SORair）的溢出率。虽然NEMA NU 4没有要求，但我们还确定了每个杆的对比度噪声比（CNRrod），表示活动恢复和图像噪声之间的权衡。对于FBP和3DRP，所应用滤波器的截止频率不同，而对于OSEM 2D和OSEM 3D，迭代次数不同。对于MAP，“平滑参数”β和均匀性约束的类型（方差或分辨率）是不同的。这些分析的结果在显示肝脏中的肿瘤的FDG注射的大鼠和显示小的皮下肿瘤和肾脏的皮质结构的F-18标记的肽注射的小鼠的图像中被证明。使用均匀方差和β =0.4的MAP获得了小直径棒CNR棒方面的最佳IQ。该设置导致RCrod，1 mm= 0.21，RCrod，2 mm=0.57，%STDunif= 1.38，SORwat= 0.0011和SORair= 0.00086。然而，使用β =0.075获得了具有非常小%STDunif的最小棒的最高活性回收率，其IQ参数分别为0.31、0.74、2.67、0.0041和0.0030。重建参数的不同设置在大鼠和小鼠图像中清楚地反映为小结构（血管、小肿瘤、肾皮质结构）的恢复与均匀身体部位（健康肝脏背景）中的图像噪声之间的权衡。Inveon PET扫描仪的最高IQ使用均匀方差的MAP重建获得。β的设置取决于特定的成像目标。(C)2010 Elsevier B. V.保留所有权利。
The image reconstruction algorithms provided with the Siemens Inveon small-animal PET scanner are filtered backprojection (FBP), 3-dimensional reprojection (3DRP), ordered subset expectation maximization in 2 or 3 dimensions (OSEM2D/3D) and maximum a posteriori (MAP) reconstruction. This study aimed at optimizing the reconstruction parameter settings with regard to image quality (IQ) as defined by the NEMA NU 4-2008 standards. The NEMA NU 4-2008 image quality phantom was used to determine image noise, expressed as percentage standard deviation in the uniform phantom region (%STDunif). activity recovery coefficients for the FOG-filled rods (RCrod), and spill-over ratios for the non-radioactive water- and air-filled phantom compartments (SORwat and SORair). Although not required by NEMA NU 4, we also determined a contrast-to-noise ratio for each rod (CNRrod), expressing the trade-off between activity recovery and image noise. For FBP and 3DRP the cut-off frequency of the applied filters, and for OSEM2D and OSEM3D, the number of iterations was varied. For MAP, the "smoothing parameter" beta and the type of uniformity constraint (variance or resolution) were varied. Results of these analyses were demonstrated in images of an FDG-injected rat showing tumours in the liver, and of a mouse injected with an F-18-labeled peptide, showing a small subcutaneous tumour and the cortex structure of the kidneys. Optimum IQ in terms of CNRrod for the small-diameter rods was obtained using MAP with uniform variance and beta=0.4. This setting led to RCrod,1 mm= 0.21, RCrod,2 mm=0.57, %STDunif= 1.38, SORwat= 0.0011, and SORair= 0.00086. However, the highest activity recovery for the smallest rods with still very small %STDunif was obtained using beta=0.075, for which these IQ parameters were 0.31, 0.74, 2.67, 0.0041, and 0.0030, respectively. The different settings of reconstruction parameters were clearly reflected in the rat and mouse images as the trade-off between the recovery of small structures (blood vessels, small tumours, kidney cortex structure) and image noise in homogeneous body parts (healthy liver background). Highest IQ for the Inveon PET scanner was obtained using MAP reconstruction with uniform variance. The setting of beta depended on the specific imaging goals. (C) 2010 Elsevier B.V. All rights reserved.