Enhancement of spatial resolution in model-based iterative CT reconstruction by using sinogram preprocessing filters

Abstract

High spatial resolution in x-ray computed tomography (CT) is paramount in many clinical applications. In conventional filtered-backprojection (FBP) reconstruction, spatial resolution is optimized by using high-resolution filter kernels that have been adjusted for decades through clinical feedback. Compared to FBP, model-based iterative reconstruction (MBIR) has shown substantial advantage in improving spatial resolution via the use of accurate forward models. As a further step in achieving the best image quality, the balance between resolution "boosting" and noise/artifact suppression needs to be carefully optimized. In maximum a posteriori probability (MAP) image estimation, a global regularization parameter is typically adjusted for this purpose. However, the single scaling parameter provides limited degree of freedom in controlling the a priori probabilities, not allowing flexible emphasis on certain desired frequency contents as realized by FBP kernels. In this study, we attempt to explore more flexible techniques for resolution enhancement in MBIR. We investigate applying a sinogram-domain frequency-boosting filter prior to performing MBIR, to provide further "boosts" in spatial resolution independent of iterative processing. Initial evaluation with a wire phantom shows that the proposed method achieves up to about 20% improvement in MTF 50% without increasing noise standard deviation. The preprocessing filter may provide a new mechanism in controlling resolution-noise tradeoff in MBIR as compared to the conventional way of adjusting a regularization parameter.