Approximate Variance Images for Penalized-Likelihood Image Reconstruction

Abstract

The authors present a fairly simple procedure for computing new approximations for the pixel variances in images reconstructed by penalized-likelihood methods. The method enables the display of variance images, which can provide an indication of uncertainty that may be helpful in medical diagnosis and in evaluation of image reconstruction algorithms. Simulations of positron emission tomography (PET) scans illustrate the accuracy of the proposed variance approximations in nonzero image pixels.