Comparison of noise equivalent count rates and image noise

The aim of this paper was to investigate the correlation between noise equivalent count (NEC) Rates and the signal-to-noise ratio (S/N) in the reconstructed images. NEC rates were determined using uniform 20- and 60-cm-tall 20-cm-diameter cylinders filled with /sup 18/F. The phantoms were scanned in both two-dimensional (2-D) and 3-D mode. The reconstructed image noise, for each frame, was estimated using the bootstrap method. For each acquired frame, 250 sinogram replicates were generated (prompts and randoms, separately), which were reconstructed using FBP and OSEM (2i8s and 4i8s). The images were filtered to a final image resolution of 6 mm. From the reconstructed image sets, mean and standard deviation images were generated, from which the average image S/N (=mean/standard deviation) was calculated within a 5-cm central ROI. The effect of applying variance reduction processing on the random events, prior to subtraction from the prompt events, on the image noise was also investigated. The NEC was found to have a linear relationship to the square of the image S/N, where the only difference found was in the slope between different reconstruction methods, implying a difference in noise properties specific to each reconstruction algorithm. The linear relationship was also found in both 2-D and 3-D acquisition modes, as well as for the different activity distributions. The randoms processing showed an improvement in image S/N, consistent with a near noiseless estimate of the random rate. These results indicate that the NEC is not only a measure for comparing the count rate performance of imaging systems, but this measure also translates into a measure for comparing image noise.