We analyzed the noise characteristics of two-dimensional (2-D) and thr
ee-dimensional (3-D) images obtained from the GE Advance positron emis
sion tomography (PET) scanner, Three phantoms were used: a uniform 20-
cm phantom, a 3-D Hoffman brain phantom, and a chest phantom with hear
t and lung inserts. Using gated acquisition, we acquired 20 statistica
lly equivalent scans of each phantom in 2-D and 3-D modes at several a
ctivity levels. From these data, we calculated pixel normalized standa
rd deviations (NSD's), scaled to phantom mean, across the replicate sc
ans, which allowed us to characterize the radial and axial distributio
ns of pixel noise. We also performed sequential measurements of the ph
antoms in 2-D and 3-D modes to measure noise (from interpixel standard
deviations) as a function of activity. To compensate for the differen
ce in axial slice width between 2-D and 3-D images (due to the septa a
nd reconstruction effects), we developed a smoothing kernel to apply t
o the 2-D data, After matching the resolution, the ratio of image-deri
ved NSD values (NSD2D/NSD3D)(2) averaged throughout the uniform phanto
m was in good agreement with the noise equivalent count (NEC) ratio (N
EC3D/NEC2D). By comparing different phantoms, we showed that the atten
uation and emission distributions influence the spatial noise distribu
tion. The estimates of pixel noise for 2-D and 3-D images produced her
e can be applied in the weighting of PET kinetic data and may be usefu
l in the design of optimal dose and scanning requirements for PET stud
ies. The accuracy of these phantom-based noise formulas should be vali
dated for any given imaging situation, particularly in 3-D, if there i
s significant activity outside the scanner field of view.