High speed algorithm for statistically reliable parametric image in positron emission tomography without continuous arterial blood sampling

基于体素的PET动力学分析可以可视化活组织中的各种功能。形成参数图像的主要缺点是基于体素的PET数据中的噪声统计较差。由于PET中存在大量的体素，计算时间过长，使得参数图像的形成变得不切实际。针对PET数据形状相似、功能相似的体素，提出了一种基于统计模型的动态聚类算法CAKS。在CAKS中，将PET数据投影到特征空间，并使用混合高斯模型。结果是，参数图像可以在30分钟内形成，并且估计值与基于顺序ROI的估计值相对应。动脉采血是将CAKS应用于临床的另一个问题。脑内组织和血管的空间分布是可区分的，适用于独立成分分析盲分离的统计算法。在专门设计了数据预处理方案以突出血管和组织空间分布的差异后，提出的算法可以估计动脉血液中活动的时间过程，并与测量的时间活动曲线相对应。

Voxel-based kinetic analysis in PET can visualize various functionalities in a living tissue. Major drawbacks in formation of an parametic image are bad noise statistics in voxel-based PET data. A large amount of voxels in PET causes a huge calculation time, and it makes formation of a parametric image inpractical. The statistical model-based clustering algorithm of CAKS (Clustering Analysis for Kinetics) is proposed to categorize voxels whose PET data has a similar shape and has similar functionalities. In CAKS, PET data was projected in a feature space and mixture Gaussian model was ulitized. As results, a parametric image can be formed in 30 minutes and the estimates were correspond to the estimates by an ordinal ROI-based estimation. Arterial blood sampling is other problem to apply CAKS in a clinical situation. The spatial distribution of tissue and blood vessel in a brain is distinguishable, and statistical algorithm for a blind separation of independent component analysis was applicable. After specialy designed data preprocessing schemes to emphasize the difference between spatial distribution of vessel and tissue, a proposed algorithm can estimate time course of activity in arterial blood and it corresponded to a measured time activity curve.