Projection data recovery induced non-local means for low-dose CT reconstruction

To improve the quality of low-dose computed tomography (CT) image, a novel projection data recovery induced non-local means for low-dose CT reconstruction is proposed. The presented method can take the advantages of data recovery methods in two domains (projection domain and image domain). Specially, the projection data is first transformed from Poisson distribution to Gaussian distribution using the nonlinear Anscombe transform in order to easily filter the noise of projection data. Second, after Anscombe transformed data is filtered, Anscombe inverse transform is performed, and the reconstructed image is achieved using the classical filtered back projection (FBP) method from filtered projection data. Last, non-local means (NL-means) weights of FBP image are computed from the restored projection data to induce the NL-means filtering of directly reconstructed FBP image from the un-restored projection data. Simulated and clinical experimental results demonstrate that the proposed method performs very well in lowering the noise and preserving the image edge.