Determination of Volume of Distribution using Likelihood Estimation in Graphical Analysis: Elimination of Estimation Bias

The graphical analysis uses an ordinary least squares (OLS) fitting of transformed data to determine the total volume of distribution (V_T) and is not dependent upon a compartmental model configuration. This method, however, suffers from a noise-dependent bias. Approaches for reducing this bias include incorporating a presmoothing step, minimizing the squared perpendicular distance to the regression line, and conducting multilinear analysis. The solution proposed by Ogden, likelihood estimation in graphical analysis (LEGA), is an estimation technique in the original (nontransformed) domain based upon standard likelihood theory that incorporates the specific assumptions made on the noise inherent in the measurements. To determine the impact of this new method upon the noise-dependent bias, we compared V_T determinations by compartmental modeling, graphical analysis (GA), and LEGA in 36 regions of interest in dynamic PET data from 25 healthy volunteers injected with [ ¹¹C]-WAY-100635 and [¹¹C]-McN-5652, which are agents used to image the serotonin IA receptor and serotonin transporter, respectively. As predicted by simulations, LEGA eliminates the noise-dependent bias associated with GA using OLS. This method is a valuable addition to the tools available for the quantification of radioligand binding data in PET and SPECT.