Medical Imaging Conference 1995 in San Francisco October 25-28.

S{\o}ren Holm, Peter Toft, and Mikael Jensen
Dept. of Nuclear Medicine, Rigshospitalet, National University
Hospital, Copenhagen, Denmark (sholm@pet.rh.dk).
Electronics Institute, Technical University of Denmark (ptoft@ei.dtu.dk).


This work determines the relative importance of noise from blank (B),
transmission (T) and emission (E) scans in PET using a GE Advance scanner on
a 20 cm cylinder, a brain phantom, and a torso-like ellipse (18/35 cm)
with examples of human scans (brain O-15 water and F-18 FDG, heart
FDG).  Phantom E scans were acquired in both 2D and 3D modes as decay
series with C-11 or F-18 over 3-6 decades of Noise Equivalent Counts
(NEC).  B and T scans were made using two pin sources (389+134 MBq)
with times 16-32768 sec. In humans only a limited subset was available.
In homogeneous phantoms normalized variance (var) was estimated from
pixel distributions in single images. In other objects, including the
human studies, calulations were performed on differences of paired
images.  In all cases a fit was made to a simple noise model. The
cylinder data shows expected relations of T to B noise proving the
adequacy of B scan times < 30 min for most purposes. For cylinder and
brain phantom, contour plots are provided for var(E,T). In a typical
3D O-15 water study with 0.5M counts per central slice, a 10 min
T-scan ensures that var(T)<0.1*var(E). Using 10 min T scan for a
static 3D FDG brain study of 10 min having 5M counts yields equal E
and T variance contributions.  In human body scans T noise has a
relative larger importance and is often dominated by effects of line
artefacts from (clusters of) zeros in the T-scan, not included in the
simple model.