TY - JOUR T1 - Extraction of the input function based on constrained NMF for dynamic <sup>18</sup>F-FDG microPET studies JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 2057 LP - 2057 VL - 52 IS - supplement 1 AU - Jie Tian AU - Xiaoqian Dai AU - Zhe Chen Y1 - 2011/05/01 UR - http://jnm.snmjournals.org/content/52/supplement_1/2057.abstract N2 - 2057 Objectives We present and validate a method to noninvasively estimate the input function from dynamic mouse 18F-FDG microPET images. The method is based on a modified constraint nonnegative matrix factorization (NMF) [1] which accounts for the spillover and partial volume effect (PVE). Methods A modified NMF method is used to decompose the dynamic PET images X as a product of two matrices: the probability distribution of tissue components matrix (the mixing matrix W) and the tissue time activity curves (TACs) matrix (the source component matrix H). We posed orthogonality constraint [2] on the mixing matrix W and sparseness constraint [3] on both matrices W and H, which can be used to correct spillover and PVE. The sparseness was controlled by a single parameter θ, which was pre-estimated by five training data sets to get an average parameter value. The gold standard input function for comparison was left ventricle (LV) derived input function (LVIF) by using 3 blood samples to correct spillover and PVE [4]. Tested microPET images were downloaded from the Mouse Quantitation Program database [5]. Twelve C57BL/6 mice data sets randomly chosen from the database were used to verify our method. Results Results showed that the derived input function was comparable in shape and magnitude to LVIF derived by [4]. Compared with TAC of LV regions of interest (LV-TAC), it was shown that the derived input function corrected the underestimation of the peak of LV-TAC by ~30% due to PVE which was consistent with [6] and also corrected the overestimation of the tail of LV-TAC due to spillover. Moreover, no significant difference was detected between the derived input function and LVIF in the myocardiac metabolic rate for glucose (Patlak Κi). Conclusions The proposed method provides a reliable, noninvasive estimation of the spillover and PVE corrected input function for dynamic small-animal 18F-FDG PET studies. Research Support This paper is supported by the National Basic Research Program of China (973 Program) under Grant No.2011CB707700, the National Natural Science Foundation of China under Grant No. 81071218 ER -