@misc{oai:repo.qst.go.jp:00069660,
 author = {Kimura, Yuichi and Shidahara, Miho and Seki, Chie and Naganawa, Mika and Sakata, Muneyuki and Ishikawa, Masatomo and Ito, Hiroshi and Kanno, Iwao and Ishiwata, Kiichi and 木村 裕一 and 志田原 美保 and 関 千江 and 長縄 美香 and 坂田 宗之 and 伊藤 浩 and 菅野 巖 and 石渡 喜一},
 month = {Dec},
 note = {Objective:  Logan graphical analysis (LGA; Logan, JCBF, 1990) is useful for neuroreceptor imaging. LGA does not assume the number of compartments to describe a behavior of administered radioligand, and it is implemented as a line estimation, that is stable and fast. However, LGA with regression line estimation suffers from an underestimation in estimated total distribution volume (VT) due to a noise in PET data (tTAC) (Kimura, ANM, 2007). In this study, a new approach combining a likelihood estimation for graphical analysis (LEGA; Ogden, Stat Med, 2003) with a maximum a posteriori (MAP) algorithm, naming MEGA (MAP estimation in graphical analysis). 
Algorithm: In LEGA, tTAC is computed from VT and an intercept in LGA (b) in the manner of LGA, and VT and b are derived in a likelihood fashion using a nonlinear estimation algorithm. In MEGA,  a set of tTACs are formed with VT and b varying in a physiological range as a prior knowledge, and then the most similar tTAC is searched for a given measured tTAC in a feature space, where a shape of tTAC is represented as a point in a multidimensional space, and a similarity between two tTACs can be evaluated using the distance as described in Kimura (NeuroImage, 1999). A noise in a measured voxel-based tTAC should be addressed. A fluctuation in a shape of tTAC is dealt with a distribution of feature points, and a Mahalanobis distance is utilized for the. Moreover, for noise reduction in a tTAC, principal components (PCs) are computed from the template to form an orthogonal subspace, and a measured tTAC is projected onto it before the searching.
Conclusion: LGA is attractive for PET neuroreceptor imaging due to its reliability and simplicity. But the noise-induced underestimation is problematic. The proposed MEGA realized robust estimation and suppressed the bias under existence of large noise observed in a voxel-based tTAC. In conclusion, MEGA can be practical algorithm for LGA neuroreceptor imaging with PET., 分子イメージング研究シンポジウム　MOLECULAR IMAGING2008},
 title = {Improvement of likelihood estimation in Logan graphical analysis using maximum a posteriori for neuroreceptor PET imaging --- toward bias-free distribution volume imaging from noise in PET data ---},
 year = {2008}
}