Comparison of ML iterative reconstruction and TV-minimization for noise reduction in CT images

This report analyzes a maximum likelihood (ML) iterative reconstruction algorithm for computed tomography in the light of some practical considerations: low dose scanning, small number of iterations, and the choice of an appropriate starting image. We show that the choice of the starting image has a great influence on early iterations, and suggest ways to use projection- and image-based total variation (TV) minimization to improve the convergence. We also argue that reconstructing with a combined projection and image TV approach is nearly equivalent to performing a full iterative reconstruction algorithm for a low dose simulation. (orig.)