Choice of initial conditions in the ML reconstruction for transmission CT with truncated projection data

Abstract

The authors investigate the effects of various initial conditions in the maximum-likelihood gradient (ML-G) iterative reconstruction of transmission map when projection data suffer from the truncation of fan beam sampling. An ML-G iteration is normally initialized with a flat initial condition (FIC)-an image with a positive constant value in each pixel, rather than a zero initial condition (ZIC)-an image with a zero value in each pixel. The authors demonstrate that using FIC in the ML iterative reconstruction can Introduce a bias to the data inside the densely sampled region (DSR), whose projection data have no truncation at every angle. To reduce this bias, the authors propose to use the largest right singular vector (LRSV) of the system matrix as initial condition, and demonstrate that this bias can be reduced with the usage of LRSV.