Development of the algorithm for quantitative neuroreceptor imaging using PET without serial arterial blood sampling and reference region

本研究旨在开发无需连续动脉采血和参考区的PET全定量神经受体成像算法。针对连续动脉采血的遗漏问题，研究了基于Logan图分析(LGA)的变分贝叶斯方法和分段搜索法。因此，在没有连续动脉采血的情况下，使用FDG、MPDX和TMSX进行了动态研究。变分贝叶斯方法将非负约束引入到基于独立分量分析的算法中，该算法是由我的团队开发的。这些方法改进了动脉血放射性时间历史(PTAC)的估计性能，并使作为该算法的另一结果的估计血容量图像在生理上可靠。基于LGA的另一种算法是使用LGA的运算方程来估计PTAC，对于这种算法，包括PTAC在内的项被用PTAC抵消的情况在区域之间很常见。由于其数学框架，该算法对PET数据中的噪声很敏感。为了实现可靠的PTAC估计，基于放射性药物给药动力学的数据预处理程序，研究了基于体素的房室模型分析的可行性。然而，由于基于体素的PET数据中的噪声统计很差，这是不实用的。然后，使用LGA估计每个体素的总分布体积(VT)，以减少估计的速率常数的数目。在PET数据中存在较大噪声的情况下，LGA估计的VT被低估。将基于MAP的VT估计方法与LGA相结合，实现了快速、可靠的基于体素的VT估计。

This study aims at developing algorithms for fully quantitative neuroreceptor imaging using PET without serial arterial blood sampling and reference regions. For omission of serial arterial blood sampling, variational Bayesian approach and an intersectional searching approach were investigated based on Logan graphical analysis (LGA). As a result, dynamic studies using FDG, MPDX and TMSX were accomplished without serial arterial blood sampling. The variational Bayesian approach introduces nonnegative constraints to an algorithm based on independent component analysis which has been developed by my group. The approaches improved an estimation performance of a time history of radioactivity in arterial blood (pTAC), and it made an estimated blood volume image, which was another outcome of the algorithm, physiologically reliable. Other algorithm based on LGA is to estimate pTAC using the operational equation of LGA, for which the terms including pTAC were canceled out using pTAC is common among regions. This algorithm was sensitive for noise in PET data due to its mathematical framework. In order to realize reliable pTAC estimation, some data preprocessing procedures were developed using clustering based on a kinetics of administered radiopharmaceutical For the omission of reference region, a feasibility of voxel-based compartment model analysis was investigated. However, it was not practical due to bad noise statistics in voxel-based PET data. Then, total distribution volumes (VT) was estimated for each voxels to decrease the number of estimated rate constants using LGA. The VT estimates by LGA is underestimated under the existence of large noise in PET data. MAP based estimation approach was incorporated with LGA, and fast and statistically reliable voxel-based VT estimation was archived.