Zeraatkar Navid, Rahmim Arman, Sarkar Saeed, Ay Mohammad Reza
Research Center for Molecular and Cellular Imaging, Tehran University of Medical Sciences, Tehran, Iran.
Department of Radiology, Johns Hopkins University, Baltimore, Maryland, US.
Asia Ocean J Nucl Med Biol. 2017 Spring;5(2):120-133. doi: 10.22038/aojnmb.2017.8708.
Various iterative reconstruction algorithms in nuclear medicine have been introduced in the last three decades. For each new imaging system, it is wise to select appropriate image reconstruction algorithms and evaluate their performance. In this study, three approaches of image reconstruction were developed for a novel desktop open-gantry SPECT system, PERSPECT, to assess their performance in terms of the quality of the resultant reconstructed images.
In the present work, a proposed image reconstruction algorithm for the PERSPECT, referred to as (qSMART), together with two popular image reconstruction methods, (MLEM) and EM (OSEM), were implemented and compared. Analytic and Monte Carlo simulations were applied for data acquisition of various phantoms including a micro-Derenzo phantom. All acquired data were reconstructed by the three algorithms using different number of iterations (1-40 ). A thorough set of figures-of-merit was utilized to quantitatively compare the generated images.
OSEM depicted reconstructed images of higher (or matching) quality in comparison to qSMART. MLEM also reached nearly similar quality as OSEM but at higher number of iterations. The graph of data discrepancy revealed that the ranking of the three approaches in terms of convergence speed is as qSMART, OSEM, and MLEM. Furthermore, bias-versus-noise curves indicated that optimal bias-noise results were achieved using OSEM.
The results showed that although qSMART can be applied for image reconstruction if being halted in the early iterations (up to 5), the best achievable quality of images is obtained using the OSEM.
在过去三十年中，核医学领域引入了各种迭代重建算法。对于每一种新的成像系统，选择合适的图像重建算法并评估其性能是明智之举。在本研究中，针对一种新型台式开放式龙门单光子发射计算机断层扫描（SPECT）系统PERSPECT开发了三种图像重建方法，以根据所得重建图像的质量评估它们的性能。
在本工作中，实现并比较了一种为PERSPECT提出的图像重建算法，称为（qSMART），以及两种流行的图像重建方法，（最大似然期望最大化算法，MLEM）和（有序子集期望最大化算法，OSEM）。应用解析模拟和蒙特卡罗模拟对包括微德伦佐体模在内的各种体模进行数据采集。所有采集的数据均通过这三种算法使用不同的迭代次数（1 - 40次）进行重建。利用一整套品质因数来定量比较生成的图像。
与qSMART相比，OSEM重建的图像质量更高（或相当）。MLEM在迭代次数较多时也达到了与OSEM几乎相似的质量。数据差异图显示，这三种方法在收敛速度方面的排名为qSMART、OSEM和MLEM。此外，偏差与噪声曲线表明，使用OSEM可获得最佳的偏差 - 噪声结果。
结果表明，虽然qSMART如果在早期迭代（最多5次）时停止可用于图像重建，但使用OSEM可获得最佳的图像质量。
