Noninvasive estimation of the input function for dynamic mouse 18F-FDG microPET studies.

Abstract

A new noninvasive estimation method for the plasma time-activity curve, i.e., input function (IF) of the tracer kinetic model in dynamic (18)F-FDG microPET mouse studies, is proposed and validated. This estimation method comprises of four steps. First, a novel constraint nonnegative matrix factorization segmentation algorithm was applied to extract the left ventricle (Lv) and myocardium (Myo) time activity curves (TACs). Second, we modeled the IF as a seven-parameter mathematical equation and constructed a dual-output model of the real TAC in Lv and Myo accounting for the partial-volume and spillover effects. Then, we fit the image-derived Lv and Myo TACs to the dual-output model to estimate the parameters of the IF. Finally, the IF was validated by comparing it to the gold standard IF while considering the delay and dispersion effects. Our method was verified based on 20 mice datasets from the Mouse Quantitation Program database, provided by UCLA. The error of the areas under the curves between the delayed and dispersed estimated IF and the gold standard IF was 7.237% ± 6.742% (r = 0.969), and the error of the (18)F-FDG influx constant Ki of the Myo was 4.910% ± 6.810% ( r = 0.992). The results demonstrated the effectiveness of the proposed method.