A. Bertoldo, G. Sparacino, C. Cobelli
Department of Information Engineering, University of Padova, Italy
Objectives: Quantification of PET images requires the adoption of a model to measure physiological parameters. Usually, parameter estimation is performed, in any given region of interest (ROI), by least squares (LS). Often, however, the signal-to-noise ratio (SNR) of PET data is, relative to model complexity, too low to allow reliable LS estimation. The goal here is to study the feasibility of a “ROI” population approach in PET.
Material & Methods: PET imaging of femoral skeletal muscle with [18F]FDG was performed in 4 healthy male volunteers. Arterial blood samples were also collected along all the experimental time. Eight ROIs were drawn on the [18F]FDG images of each subject in the anteromedial muscle compartments of the femoral region, carefully avoiding the great vessels. The four-compartment five-rate-constant (5K) model developed for describing [18F]FDG kinetics in human skeletal muscle was used for image quantification. For each subject, parameter estimates of 5K model were obtained in each ROI both by LS and by nonlinear mixed-effects modeling with a ROI population described by the 8 drawn ROIs. The Iterative Two Stage method was also considered as a possible computationally attractive alternative to nonlinear mixed-effects modeling. The mixed-effects modeling was performed using NONMEM. The ITS algorithm was implemented in Matlab6.
Results: By using LS, positive parameter estimates and “reasonable” CV (i.e. <500%) were obtained in 6 of the 8 ROIs for the first two subjects and only in 1 and 4 of the original 8 ROIs in the remaining two subjects. By using NONMEM-FOCE, successful convergence of parameter estimation was achieved in all ROIs. NONMEM-FOCE allows the calculation of individual estimates with precision comparable or better than that obtained using LS also when LS parameter precision is unacceptable. It is of note that NONMEM-FOCE estimated fixed-effects describe an interROI variability very different from that obtainable using the simple mean and variance of LS estimates. With respect to LS and mixed-effects modeling results, ITS allows the calculation of more precise individual estimates for all the ROIs in all the subjects. ITS algorithm convergence was achieved after few iterations. We did not detect any significant difference (P<0.05) between individual NONMEM and ITS estimates.
Conclusion: The results show that the use of population approaches allow a more accurate and more precise determination of parameters of kinetic models from PET data, even in the presence of highly noisy images. In particular, ITS has the potential to allow a reliable and computationally less intensive than NONMEM quantification of the images.
Reference: PAGE 12 (2003) Abstr 440 [www.page-meeting.org/?abstract=440]
Poster: poster