Rong Fu Wang,
Peking University Health Science Center, China
Title: Generalization and optimization of a population-based input function estimation and approach for quantification of sparsely sampled dynamic 18F-FDG PET/CT data
Biography
Biography: Rong Fu Wang,
Abstract
18F-FDG uptake rate constant Ki is a most interested and commonly used parameter for absolute quantification of 18F-FDG
PET/CT. Ki is usually estimated by fitting a kinetic model with plasma input function (PIF) to the measured dynamic PET
data. The need for arterial blood sampling to measure PIF (mPIF) is a main barrier to estimate Ki for clinical 18F-FDG PET.
Two existing noninvasive PIF estimation methods, image derived PIF and simultaneous fitting method with kinetic model and
parametric PIF, require image data to be acquired continuously and immediately post tracer injection. The objective of the study
is to validate and optimize a generalized population-based PIF estimation method for noninvasive quantification of dynamic
18F-FDG PET for sparsely sampled PIF. Eight 60-min 27-frame monkey dynamic 18F-FDG PET studies were collected from a
Philips Gemini GXL PET/CT with 3D data acquisition mode. Total 34 arterial blood samples were taken during PET scan as:
22 samples for the first 4 min, and followed by sampling at 5, 6, 8, 10, 12, 15, 20, 25, 30, 40, 50 and 60 min. Time activity curves
(TACs) of 7 cerebral regions of interests (ROIs) were generated from each study. A generalized population-based approach to
recover full kinetics of the PIF from sparsely sampled blood data is proposed. The estimated PIF (ePIF) from the incomplete
PIF sampling data was determined by interpolation and extrapolation using scale-calibrated population mean of normalized
PIFs. The optimal blood sampling scheme with given sample size m was determined by maximizing coefficient coefficients
of Ki estimates between the Kis from measured PIF (mPIF) and estimated PIF (ePIF). The leave-two-out cross validation was
performed. The linear correlations between the Ki estimates from the ePIF (with optimal sampling scheme) and those from
the mPIF were: Ki(ePIF; 1 sample at 40 min) = 1.015Ki(mPIF) -0.000, R2 = 0.974; Ki(ePIF; 2 samples at 25 and 50 min) =
1.029Ki(mPIF) - 0.000, R2 = 0.985; Ki(ePIF; 3 samples at 8, 20, and 50 min) = 1.039Ki(mPIF) - 0.001, R2= 0.993; and Ki(ePIF;
4 samples at 8,12, 25, 40, and 55 min) = 1.02Ki(mPIF)-0.000, R2=0.997. The correlations of R2 from leave-2-out validation
study were 0.978±0.007, 0.990±0.006, and 0.996±0.003 (mean ±SD) for 1, 2, and 3 samples of optimal sampling scheme,
respectively. The generalized population-based PIF estimation method is a reliable method to estimate PIFs from incomplete
blood sampling data for quantification of dynamic 18F-FDG PET using the Gjedde-Patlak plot.