Evaluating popular non-linear image processing filters for their use in regularized iterative CT

Abstract

Abstract-Iterative CT algorithms are becoming increasingly popular in recent years, and have been found useful when the projections are limited in number, irregularly spaced, or noisy, which are imaging scenarios often encountered in low-dose imaging and compressed sensing. One way to cope with the associated streak and noise artifacts in these settings is either to incorporate or to interleave a regularization objective into the iterative reconstruction framework. In this paper we explore possible techniques for the latter. We investigate a number of non-linear filters popular in the image processing literature for their suitability in iterative CT application, here OS-SIRT.

GPU-based high-performance computing for radiation therapy

“…Later, the full reconstruction process was implemented on GPU. Specifically, Xu and Mueller (2010) inserted the TV minimization in each iteration of their OS-SART loop. Jia et al treated the reconstruction as an optimization problem in which the objective function contained both a least-square term to enforce the projection condition and a TV term to regularize the image (Jia et al 2010a(Jia et al , 2011c.…”

“…This filter was used by Xu et al in their reconstruction framework together with the OS-SART algorithm (Xu andMueller 2009, 2010). A more general form of this filter, non-local-means was also utilized by the same group (Xu and Mueller 2010), where the weighting factors were obtained by comparing patches centering at each voxel, rather than the voxels themselves. Finally, another type of regularization method was invented based on the assumption that the reconstructed image has a sparse representation under the tight wavelet-frame basis (Jia et al 2011a).…”