Statistical cone-beam CT noise reduction with multiscale decomposition and penalized weighted least squares in the...

Suppressing noise can effectively promote image quality and save radiation dose in clinical imaging with x-ray computed tomography (CT). To date, numerous statistical noise reduction approaches have ever been proposed in image domain, projection domain or both domains. Especially, a multiscale decomposition strategy can be exploited to enhance the performance of noise suppression while preserving image sharpness. Recognizing the inherent advantage of noise suppression in the projection domain, we have previously proposed a projection domain multiscale penalized weighted least squares (PWLS) method for fan-beam CT imaging, wherein the sampling intervals are explicitly taken into account for the possible variation of sampling rates. In this work, we extend our previous method into cone-beam (CB) CT imaging, which is more relevant to practical imaging applications.

The projection domain multiscale PWLS method is derived for CBCT imaging by converting an isotropic diffusion partial differential equation (PDE) in the three-dimensional (3D) image domain into its counterpart in the CB projection domain. With adoption of the Markov random field (MRF) objective function, the CB projection domain multiscale PWLS method suppresses noise at each scale. The performance of the proposed method for statistical noise reduction in CBCT imaging is experimentally evaluated and verified using the projection data acquired by an actual micro-CT scanner.

The preliminary result shows that the proposed CB projection domain multiscale PWLS method outperforms the CB projection domain single-scale PWLS, the 3D image domain discriminative feature representation (DFR), and the 3D image domain multiscale nonlinear diffusion methods in noise reduction. Moreover, the proposed method can preserve image sharpness effectively while avoiding generation of novel artifacts.

Since the sampling intervals are explicitly taken into account in the projection domain multiscale decomposition, the proposed method would be beneficial to advanced applications where the CBCT imaging is employed and the sampling rates vary.