Properties Of Statistical Noise In Positron Emission Tomography

The magnitude and texture of statistical noise in images of positron emission computed tomography are evaluated analytically. A simple approximation method is presented for evaluating the variance and autocovariance function of noise for objects having given distributions of radionuclides and of attenuation coefficient. The comparison between the approximation method and the accurate estimation shows excellent agreements at various points in the images of uniform disc sources having constant attenuation. The dependence of noise amplification on convolution filters in image reconstruction and on the additional image processing (smoothing) is given. Considerable anisotropy of the autocovariance is observed near the periphery of the image even for a uniform disc source. This fact suggests the usefulness of spatially variant, anisotropic smoothing in some cases.