Yan Jianhua, Lim Jason Chu-Shern, Townsend David W
A*STAR-NUS Clinical Imaging Research Center, 14 Medical Drive, #B1-01, 117599, Singapore.
Phys Med Biol. 2015 Feb 7;60(3):961-76. doi: 10.1088/0031-9155/60/3/961. Epub 2015 Jan 9.
Positron emission tomography (PET) image quantification is a challenging problem due to limited spatial resolution of acquired data and the resulting partial volume effects (PVE), which depend on the size of the structure studied in relation to the spatial resolution and which may lead to over or underestimation of the true tissue tracer concentration. In addition, it is usually necessary to perform image smoothing either during image reconstruction or afterwards to achieve a reasonable signal-to-noise ratio. Typically, an isotropic Gaussian filtering (GF) is used for this purpose. However, the noise suppression is at the cost of deteriorating spatial resolution. As hybrid imaging devices such as PET/MRI have become available, the complementary information derived from high definition morphologic images could be used to improve the quality of PET images. In this study, first of all, we propose an MRI-guided PET filtering method by adapting a recently proposed local linear model and then incorporate PVE into the model to get a new partial volume correction (PVC) method without parcellation of MRI. In addition, both the new filtering and PVC are voxel-wise non-iterative methods. The performance of the proposed methods were investigated with simulated dynamic FDG brain dataset and (18)F-FDG brain data of a cervical cancer patient acquired with a simultaneous hybrid PET/MR scanner. The initial simulation results demonstrated that MRI-guided PET image filtering can produce less noisy images than traditional GF and bias and coefficient of variation can be further reduced by MRI-guided PET PVC. Moreover, structures can be much better delineated in MRI-guided PET PVC for real brain data.
正电子发射断层扫描（PET）图像定量是一个具有挑战性的问题，这是由于采集数据的空间分辨率有限以及由此产生的部分容积效应（PVE），部分容积效应取决于所研究结构的大小与空间分辨率的关系，并且可能导致对真实组织示踪剂浓度的高估或低估。此外，通常需要在图像重建期间或之后进行图像平滑处理，以获得合理的信噪比。通常，为此目的使用各向同性高斯滤波（GF）。然而，噪声抑制是以牺牲空间分辨率为代价的。随着PET/MRI等混合成像设备的出现，从高分辨率形态图像中获得的互补信息可用于提高PET图像的质量。在本研究中，首先，我们通过采用最近提出的局部线性模型提出一种MRI引导的PET滤波方法，然后将PVE纳入该模型，以获得一种无需对MRI进行分割的新的部分容积校正（PVC）方法。此外，新的滤波和PVC都是体素级非迭代方法。我们使用模拟的动态FDG脑数据集以及用同步混合PET/MR扫描仪采集的一名宫颈癌患者的（18）F-FDG脑数据，对所提出方法的性能进行了研究。初步模拟结果表明，MRI引导的PET图像滤波可以产生比传统GF噪声更少的图像，并且通过MRI引导的PET PVC可以进一步降低偏差和变异系数。此外，对于真实脑数据，在MRI引导的PET PVC中可以更好地描绘结构。
