Limiting iterations vs. post smoothing for noise control in PET

We have performed physical phantom and simulation study with PET to compare the effects of limiting iterations and post-smoothing. The phantom was designed to mimic small hot lesions typical in FDG PET. The phantom was a body-sized 25 L phantom. Eight 1 cm hot spheres were filled with activity 8x the background. The phantom was scanned in 2D for 15 minutes on a GE Discovery STE. Data were framed into 1, 3, 5, and 15 min frames. Images were reconstructed with a variety of subset sizes and up to 60 iterations, with corrections for attenuation, scatter, randoms, and dead-time. All image sets were post-smoothed with 4, 6, 8, 10, and 12 mm FWHM Gaussian. ROI's were placed on all hot spheres, and 40 additional background ROI's were placed. Noise was measured as the variation in background ROI's. Additionally, noise was measured on a pixel-by-pixel basis as the variation over the ensemble of realizations of each frame time. Noise vs. contrast (sphere-background) were plotted for each image. The background noise method showed better results for limited iterations (~20% lower noise at a given contrast level) while the ensemble-based noise method favored post-smoothing of high iterations (with a similar ~20% difference). Similar results were obtained for simulations.