Adaptive filtering with self-similarity for low-dose CT imaging

An important problem in low-dose CT is the image quality degradation caused by photon starvation. There are a lot of algorithms in sinogram domain or/and image domain to try to overcome this difficulty. In view of strong self-similarity contained in the special sinusoid-like strip data in the sinogram space, we propose an adaptive filtering with self-similarity, whose average weights are related to both the image FBP (filtered backprojection) reconstructed from the restored sinogram data and the image directly FBP reconstructed from the noisy sinogram data in a framework of weighted average processing. In order to filter sinogram data, a non-local means method is used with its smoothing adaptive to the variances of noisy data after an adaptive median filtering, which preserves important features and high accuracy of the data in sinogram domain. In simulation experiments, it is shown that our proposed method, with filtering in both image and projection domains, has a better performance in noise reduction and feature preservation in reconstructed images.