Sparse-View Computed Tomography Reconstruction Using an Improved Non-Local Means

Abstract

In X-ray computed tomography (CT) examinations nowadays, radiation dose reduction without degrading CT images has caused significant concerns. One simple and effective way for dose reduction is to reduce the number of X-ray projections to reconstruct CT images. Non-local means (NLM) based reconstruction methods have been studied for years, but often lead to over-smoothness on edge information in a reconstructed image. In this work, an adaptive NLM (ANLM) into sparse-projection image reconstruction was introduced, named as ART-ANLM. For ANLM, a novel similarity measure that is rotationally invariant between any two patches and a dynamic filter parameter were proposed to solve the problem from NLM. The ART-ANLM algorithm was validated on digital and real projection data. Results have demonstrated that the proposed method could achieve a good compromise between noise suppression and structure information preserving, compared to other existing reconstruction methods.