Projection data restoration guided non-local means for low-dose computed tomography reconstruction

To improve the CT image quality from low-dose projection data measurement, a projection data restoration guided nonlocal means (NL-means) method was proposed, which makes fully use of the advantages of both the low-dose CT projection data restoration and image domain edge-preserving filtering. To get an ideal FBP reconstructed image, the projection data was first restored using Anscombe transform and block-matching and 3D filtering (BM3D) method according to the Poisson statistical nature and noise properties of the projection data. To preserve edge information during noise reduction in image domain, we proposed the optimized NL-means weights using the estimated ideal FBP image. Simulated experimental results demonstrate that the presented method performs better than other popular methods in lowering the noise and preserving the image edge, without noticeable sacrifice of the spatial resolution.