Penalized weighted alpha-divergence approach to sinogram restoration for low-dose X-ray computed tomography

Abstract

In X-ray computed tomography (CT) reconstruction, accurate statistical modeling of the measurement after linearity/nonlinearity calibration is essential to yield high quality diagnostic images, especially for low-dose CT. Due to the complicate noise distribution of after-log projection data or sinogram data, direct image reconstruction by filtered backprojection (FBP) approach is a very challenging task for noise reduction. As studied in our previous work, a (alpha) divergence as an important information metric has shown its advantages in describing the statistical distribution of sinogram data. In practice, the estimation of sinogram data is relevant to each detector bin, and the mismatched measure between the estimated and measured sinogram data should be balanced by using data-dependent weight factor at different detector bins, such as weighted least-square approach. With above observations, based on our previous work, in this paper, we propose a penalized weighted alpha-divergence (PWAD) approach for low-dose (i.e., low-mAs) CT sinogram iterative restoration. To test the performance of the present PWAD approach, a modified digital Shepp-Logan phantom and a physical phantom were used in our study. The results show that the present PWAD approach could significantly reduce the noise with less sacrificing image resolution. As a conclusion, the weighted alpha-divergence metric may be an interesting choice for building more reasonable cost-function in low-dose CT image reconstruction.