Extraction of input function from 18FDG-PET images

Serial arterial blood sampling is required to determine the input function (IF) in quantitative estimation of physiological parameters in dynamic positron emission tomography (PET) studies. However, blood sampling in PET studies is problematic, invasive and risky. Several approaches have been proposed to extract IF from images among them factor analysis of dynamic structures (FADS) and simultaneous estimation of IF and physiological parameters. FADS, needs a dimension reduction of data, intermediate steps as oblique analysis, and simultaneous estimation needs to fit exponential parameters. In this paper, we introduced a new and simple approach based on the decomposition of image pixel intensity in blood and tissue using Bayesian predictions. The method used an a priori knowledge of the probable distribution of blood and tissue in the images. Weights were then calculated accounting for concentrations of the radiotracer in blood and tissue and their contribution in each image pixel. The calculated images provided more accurate blood and tissue components free from spillover from each other. The results show that this new approach generated tissue glucose metabolic values comparable to those obtained with manual blood sampling (p = 4.13), while these values were significantly different from those calculated with region of interest drawn on the blood pool (p = 2.6.10^-4). The new approach is fast and simple in implementation and directly generates predefined components with homogeneous intensities without the recourse to fitting parameters or non-negativity constraints. The method can also be used to generate images of desired components.