SUMMARY

Computed tomography (CT) imaging dose becomes an increasing public concern nowadays. Iterative reconstruction (IR) algorithms have shown promise on noise reduction of CT images reconstructed from projection data with reduced radiation by modeling the physical process of a CT scan and incorporating prior knowledge. However, IR algorithms are criticized for its less visually appealing and over-smoothed noise texture compared with that of the conventional filtered-back-projection (FBP) method, and little research has been devoted to studies on such noise statistical properties of IR algorithms. In this study, we propose a novel iterative reconstruction method that preserves a similar noise texture with that of FBP as well as reduces image noise. The noise reduction in this novel method is achieved by reducing the variation of the pixels belonging to same material, instead of regularizing the local spatial variation of image as existing IR methods do. Secondly, we develop a practical method for pixel-wise estimation of noise statistics on clinical IR CT images, which enables accurate quantification of noise variance and noise texture for general IR methods.