An ordered subsets expectation maximization (OS-EM) algorithm is used for i
mage reconstruction to suppress image noise and to make non-negative value
images. We have applied OS-EM to a digital brain phantom and to human brain
F-18-FDG PET kinetic studies to generate parametric images. A 45 min dynam
ic scan was performed starting injection of FDG with a 2D PET scanner. The
images were reconstructed with OS-EM (6 iterations, 16 subsets) and with fi
ltered backprojection (FBP), and K1, k2 and k3 images were created by the M
arquardt non-linear least squares method based on the 3-parameter kinetic m
odel. Although the OS-EM activity images correlated fairly well with those
obtained by FBP, the pixel correlations were poor for the k2 and k3 paramet
ric images, but the plots were scattered along the line of identity and the
mean values for K1, k2 and k3 obtained by OS-EM were almost equal to those
by FBP. The kinetic fitting error for OS-EM was no smaller than that for F
BP. The results suggest that OS-EM is not necessarily superior to FBP for c
reating, parametric images.