Noise Removal of Low-Dose CT Images Using Modified Smooth Patch Ordering

Low-dose computed tomography (LDCT) images tend to be severely degraded by excessive mottle noise and steak artifacts. In this paper, an algorithm of modified smooth patch ordering (MSPO) is proposed to improve the LDCT images. In the MSPO method, the non-local means (NLM) algorithm is modified by replacing the Leclerc robust function with the modified bisquare robust function, to serve as weight function for the estimate of each pixel value. Then, the modified NLM algorithm is combined with smooth ordering of the pixels, patch classification, and subimage averaging scheme to denoise the LDCT image. Additionally, the prewhitening of the LDCT image is carried out to enhance image denoising, and the total-variation filter is utilized as a post-processing step to further remove the residual noise of the recovered image. Subjective and objective evaluations on the actual thoracic phantom and clinical data are carried out for validating the effectiveness of the proposed method. The results from computer experiments demonstrate that the proposed MSPO approach performs better in both artifact suppression and structure preservation, when compared with several existing methods. Especially, the MSPO approach can be directly applied to process digital imaging and communications in medicine (DICOM) images, and has great potential in most current CT systems.