目的：（1）研究不同采集条件，即采集时间、注射剂量、和床位重叠，对PET图像质量和定量精度的影响。（2）研究不同重建条件，即PSF建模与TOF技术、图像更新次数（迭代次数×子集）、重建矩阵、高斯后滤波核宽度，对PET图像质量和定量精度的影响。（3）研究新一代SiPM PET/CT与传统PMT PET/CT空间分辨力及图像质量，探讨不同的PET光电转换器和Q.Clear算法对PET空间分辨力、定量精度和图像质量的影响。

资料与方法：采用GE Discovery Elite型PET/CT和GE Discovery MI型PET/CT，使用不同采集与重建条件对NEMA NU2-2018图像质量 (image quality, IQ) 模体成像。（1）采集条件：①采集时间为1.5 min、2.5 min、3.5 min、6 min、15 min及30 min；②IQ模体背景活度为5.24 kBq/ml的3/2、1、1/2、1/4、1/6及1/8倍，均使用采集时间为2 min、6 min、30 min以及等计数方式采集；③两床位采集，3 min每床位，床位重叠为5层～35层，间隔5层。（2）重建条件：①使用不同重建算法：OSEM、OSEM+PSF、OSEM+TOF、OSEM+TOF+PSF、FBP；②使用不同迭代次数与子集组合，图像更新次数为16次～136次；③重建矩阵192×192、256×256和384×384，每个重建矩阵在6 min采集时间条件下分别使用30 cm、50 cm和70 cm的DFOV进行重建，在DFOV 50 cm条件下采集时间分别使用1.5 min、6 min和30 min；④高斯后滤波核宽度0 mm～8 mm，间隔2 mm。（3）不同PET/CT对比：使用椭圆分辨力模体以及IQ模体对比Elite型PET/CT和MI型PET/CT空间分辨力、定量精度和图像质量，Q.Clear算法β取150～550，间隔100。GE Xeleris工作站用于小球感兴趣区 (region of interest, ROI) 的勾画与数据提取，Matlab用于背景ROI和肺插件ROI的勾画与数据提取以及空间分辨力半高全宽 (full width at half maximum, FWHM) 的计算。使用恢复系数 (recovery coefficient, RC) 评价定量精度；对比度恢复系数 (contrast recovery coefficient, CRC)、背景变异百分比 (percent background variability, PBV)、背景变异系数 (background coefficient of variance, BCV)、对比噪声比 (contrast to noise ratio) 评价图像质量；肺插件残余误差 (residual error, RE) 评价衰减与散射校正精度。

结果：（1）延长采集时间、增加注射剂量和床位重叠均提高PET定量精度及图像质量。采集时间≥3.5 min可以使定量精度和图像质量达到最优，缩短采集时间对10 mm小球RC影响最大，采集时间为1.5 min时小球边缘出现异常放射性浓聚。散射分数随注射剂量的减少增大；当采集时间为2 min时，注射剂量降低至5.3 kBq/ml的1/2仍能获得较好的定量精度与图像质量；剩余剂量由1/4降低至1/8时10 mm小球RC先降低后增加；等计数采集方法得到的定量精度与图像质量在不同活度水平中具有较好一致性。床位重叠的选择以21%～35%为宜，床位重叠减小至5层10 mm小球难以视及，床位衔接处产生明显伪影。（2）PSF建模与TOF技术联用提高小球RC，PSF建模改善图像背景噪声与CNR，TOF技术明显降低肺插件RE。图像更新次数的增加提高小球RC与图像噪声，降低目标物探测能力，≥51次时RC变化趋于稳定；少子集多次迭代实现特定图像更新次数降低了BCV。采集时间≥6 min时增大重建矩阵提高小球RC和CNR，背景噪声受矩阵影响较小；减小DFOV提高了小球CNR。高斯滤波核宽以4 mm为宜，增大高斯滤波核宽降低图像噪声和RC，提高小球CNR和肺插件RE。（3）MI与Elite相比，空间分辨力FWHM与背景噪声减小。Q.Clear与OSEM相比改变了空间分辨力的空间分布，得到的CNR、RE和FWHM均随β增大而增大，RC、CRC、PBV和BCV均随β增大而减小。

结论：在临床成像条件下，采集与重建条件的变化在不同程度上影响PET定量精度与图像质量，不同参数的选择需要在PET定量精度、图像质量与临床可行性间权衡。（1）采集时间≥3.5 min时定量精度和图像质量达到最优，采集时间小于3.5 min时应尤其注意对10 mm及以下病灶定量指标的解释；注射剂量降低至临床常用剂量的一半能保持较好的定量精度与图像质量；床位重叠的选择以21%～35%为宜。（2）PSF建模改善图像噪声与探测能力，TOF技术提高衰减散射校正精度，两者联用提高定量精度；图像更新次数为51次、高斯滤波核宽为4 mm时能够较好地权衡定量精度与图像质量；采集时间≤6 min时重建矩阵对定量精度和图像质量影响较小，减小DFOV有助于提高目标物探测能力。（3）新一代SiPM PET和Q.Clear算法的使用提高PET空间分辨力、定量精度和图像质量，β值影响Q.Clear算法的收益。

Objective: (1) To investigate the effects of different acquisition conditions, i.e., acquisition time, injection dose, and bed overlap, on PET image quality and quantitative accuracy. (2) To investigate the effects of different reconstruction conditions, i.e., PSF modeling and TOF technique, number of image updates (iteration×subset), reconstruction matrix, and Gaussian post-filter kernel width, on PET image quality and quantitative accuracy. (3) To investigate the spatial resolution and image quality of new-generation SiPM PET/CT and traditional PMT PET/CT, and to explore the effects of different PET photoelectric transducer and Q.Clear algorithm on PET spatial resolution, quantitative accuracy and image quality.

Materials and methods:  GE Discovery Elite PET/CT and GE Discovery MI PET/CT systems were employed to perform NEMA NU2-2018 image quality (IQ) phantom imaging under various acquisition and reconstruction conditions. (1) Acquisition conditions: ① Scan durations were set to 1.5, 2.5, 3.5, 6, 15, and 30 minutes. ② IQ phantom background activity was varied at six levels: 3/2, 1, 1/2, 1/4, 1/6 and 1/8 of 5.24 kBq/ml. These variations were investigated with scan durations of 2 minutes, 6 minutes, 30 minutes and equivalent counting acquisition method. ③ Two-bed acquisition was performed with a 3-minute acquisition time per bed and overlap ranging from 5 to 35 layers with 5-layer intervals. (2) Reconstruction conditions:  ① Different reconstruction algorithms were used including OSEM, OSEM+PSF, OSEM+TOF, OSEM+TOF+PSF, FBP. ② Different combinations of iteration and subset were employed, with image update ranging from 16 to 136 times. ③ Three reconstruction matrix were used including 192×192, 256×256, and 384×384. Each matrix was reconstructed using  a 6-minute scan duration with DFOV of 30 cm, 50 cm, and 70 cm, respectively. Furthermore, scan durations of 1.5 min, 6 min, and 30 min were employed with a DFOV of 50 cm. ④Gaussian post-filtering kernels width ranging from 0 to 8 mm were used at 2 mm intervals. (3) Comparisons between Different PET/CT: The spatial resolution, quantitative accuracy, and image quality of Elite PET/CT and MI PET/CT were compared using an elliptical resolution phantom and an IQ phantom. The Q.Clear algorithm was implemented with β value ranging from 150 to 550, with intervals of 100. The GE Xeleris workstation was used to define regions of interest (ROI) for the spheres and extract data. Matlab was used to define regions of interest (ROIs) for the background and lung insert, extract data, and calculate the full width at half maximum (FWHM) for spatial resolution. The recovery coefficient (RC) was used to evaluate quantitative accuracy. Image quality was assessed using several metrics, including contrast recovery coefficient (CRC), percent background variability (PBV), background coefficient of variance (BCV), and contrast to noise ratio (CNR). The accuracy of attenuation and scatter correction was evaluated using the residual error (RE) of the lung insert.

Results: (1) PET quantitative accuracy and image quality were improved by prolonging the scan duration, increasing the injection dose, and overlap between the beds. Quantitative accuracy and image quality were optimized with an acquisition time of at least 3.5 minutes. Shortening the acquisition time can negatively impact the RC of 10 mm spheres, resulting in anomalous radioactivity concentration at the edges of the spheres. The scatter fraction increased as the injected dose decreased. Even when the injected dose was reduced to half of 5.3 kBq/ml and the acquisition time was 2 minutes, good quantitative accuracy and image quality can still be achieved. The RC of 10 mm spheres initially decreases and then increases when the remaining dose is reduced from 1/4 to 1/8. Equivalent counting methods demonstrated good consistency in quantitative accuracy and image quality across different activity levels. It was advisable to have a bed overlap of 21% to 35%. Reducing bed overlap to 5 layers makes it difficult to discern 10 mm spheres, and noticeable artifacts occurred at the bed junctions. (2) The RC of spheres was improved by the combination of PSF modeling and TOF, while PSF modeling improved image background noise and CNR. The RE of lung insert was significantly reduced by TOF. Increasing the number of image update enhanced the RC of spheres but also increased image noise, while reducing target detection capability. RC stabilized beyond 51 iterations. BCV was reduced by employing fewer subsets with multiple iterations to achieve a specific number of image updates.  The RC and CNR of spheres were improved by using a scan duration of no less than 6 minutes and increasing the reconstruction matrix size. The background noise was less affected by the matrix. The sphere's CNR increased as the DFOV decreased. It is optimal to use a Gaussian filter with a width of 4 mm. By increasing the width of the Gaussian filter, the image noise and the RC of the sphere were decreased, while the CNR of the spheres and the RE were increased. (3) Compared to Elite, MI shows reduced spatial resolution FWHM and background noise. Compared to OSEM, Q.Clear changes the spatial distribution of spatial resolution, resulting in increased CNR, RE, and FWHM with increasing β values, while RC, CRC, PBV, and BCV decrease with increasing β values.

Conclusions: Variations in acquisition and reconstruction parameters can affect PET quantitative accuracy and image quality to varying degrees under clinical imaging conditions. The selection of different parameters requires a balance between PET quantitative accuracy, image quality, and clinical feasibility. (1) When the acquisition time is longer than 3.5 minutes, both the quantitative accuracy and the image quality are at their best. However, for acquisition times less than 3.5 minutes, it is important to carefully interpret quantitative measures for lesions less than or equal to 10 mm. Halving the administered dose to commonly used clinical doses can maintain accurate quantification and image quality. Finally, it is recommended to aim for a bed overlap between 21% and 35%. (2) PSF modeling improves image noise and detectability, while TOF technology enhances attenuation and scatter correction accuracy. The combination of both improves quantitative accuracy. A balance between quantitative accuracy and image quality can be achieved with 51 iterations and a Gaussian filter kernel width of 4 mm. The impact of the reconstruction matrix on quantitative accuracy and image quality is minimal when the acquisition time is 6 minutes or less. Reducing the DFOV can improve detectability. (3) The utilization of next-generation SiPM PET and the Q.Clear algorithm enhances PET spatial resolution, quantitative accuracy, and image quality. The choice of β value affects the benefits of the Q.Clear algorithm.