A Spectral CT Denoising Algorithm based on Weighted Block Matching 3D Filtering

In spectral CT, an energy-resolving detector is capable of counting the number of received photons in different energy channels with appropriate post-processing steps. Because the received photon number in each energy channel is low in practice, the generated projections suffer from low signal-to-noise ratio. This poses a challenge to perform image reconstruction of spectral CT. Because the reconstructed multi-channel images are for the same object but in different energies, there is a high correlation among these images and one can make full use of this redundant information. In this work, we propose a weighted block-matching and three-dimensional (3-D) filtering (BM3D) based method for spectral CT denoising. It is based on denoising of small 3-D data arrays formed by grouping similar 2-D blocks from the whole 3-D data image. This method consists of the following two steps. First, a 2-D image is obtained using the filtered backprojection (FBP) in each energy channel. Second, the proposed weighted BM3D filtering is performed. It not only uses the spatial correlation within each channel image but also exploits the spectral correlation among the channel images. The proposed method is evaluated on both numerical simulation and realistic preclinical datasets, and its merits are demonstrated by the promising results.