Streak artifact reduction in low dose computed tomography

Accurate noise model is essential for dose reduction in iterative reconstruction (IR) in computed tomography, and has been studied extensively in literature. It is also important to understand how the noise model can be used for noise reduction at low counts without sacrificing spatial resolution and image quality. In this work we present a method to reduce data noise caused by a low photon count and include statistical information into reconstruction algorithm. Our approach is based on computing noise model tables for each system. Our extensive evaluation with experimental data shows that analytic noise model very closely describes real data noise, provided we have accurately measured noise model tables. Noise model tables are used (1) for before-log data pre-processing, and (2) to compute statistical weight for weighted OS-SART algorithm. Evaluation results show effectiveness of our approach and potential for substantial dose reduction.