Image Noise Variance in 3D OSEM Reconstruction of Clinical Time-of-Flight PET

We have investigated image noise variance in 3D OSEM reconstruction of clinical time-of-flight (TOF) and conventional PET. Patient data were acquired on a biograph 16 PET/CT having 1.25 nsec TOF time resolution. Statistically independent replicate sinograms obtained from sampled listmode data were analyzed. We computed TOF and non-TOF image variance as a function of iteration number and scan time, extrapolated to convergence, and determined the dependence of both converged and post-smoothed image variances on scan duration. We have found that these image variances have an inverse power-law dependence on scan time, but with exponents < 1, likely due to the noise-dependent bias of OSEM. This means that scan time must be increased by more than a factor f in order to reduce image variance by 1/f, and hence that image variance is not inversely proportional to NEC. Image noise variance is found to be 38-48% higher for non-TOF compared to TOF, for equal scan time. For equal variance, conventional reconstruction requires 63-73% longer scan time than does TOF. This scan time advantage is similar to what we have observed in linear reconstruction, and indicates that TOF has a distinct benefit for clinical iterative reconstruction even with 1.25 nsec time resolution.