Employing temporal self-similarity across the entire time domain in computed tomography reconstruction

Abstract

There are many cases where one needs to limit the X-ray dose, or the number of projections, or both, for high frame rate (fast) imaging. Normally, it improves temporal resolution but reduces the spatial resolution of the reconstructed data. Fortunately, the redundancy of information in the temporal domain can be employed to improve spatial resolution. In this paper, we propose a novel regularizer for iterative reconstruction of time-lapse computed tomography. The non-local penalty term is driven by the available prior information and employs all available temporal data to improve the spatial resolution of each individual time frame. A high-resolution prior image from the same or a different imaging modality is used to enhance edges which remain stationary throughout the acquisition time while dynamic features tend to be regularized spatially. Effective computational performance together with robust improvement in spatial and temporal resolution makes the proposed method a competitive tool to state-of-the-art techniques.

Recent developments in neutron imaging with applications for porous media research

“…In the fluid-flow problems of interest here, the initial stationary "dry" stage is an ideal "prior" reference for iterative reconstruction. Simultaneous iterative reconstruction (SIRT) and the conjugate gradient least square (CGLS) methods employed are now available (Van Eyndhoven et al, 2015;Kazantsev et al, 2015a). Here we summarize the findings that are of specific interest to porous media experiments.…”

An Iterative CT Reconstruction Algorithm for Fast Fluid Flow Imaging

“…The reader is referred to [19] for details about the method and its parameters. The prior image containing structural information was set to be the same static reconstruction that was utilized in the method by Meyers et al Most algorithm parameters were adapted from [19], except for the number of iterations, the regularization parameter and the noise-dependent parameter (denoted by MaxOuter, β and h in [19], respectively), since these parameters are problem dependent. These three parameters were optimized for lowest RRMSE with respect to the ground truth for each and every experiment in the simulation experiments and selected manually based on visual assessment in the neutron tomography experiment.…”

“…• CGLS-NLST: The Conjugate Gradient Least Squares method with Non-Local Spatio-Temporal penalty (CGLS-NLST) developed by Kazantsev et al and implemented as described in [19]. The reader is referred to [19] for details about the method and its parameters.…”

“…These three parameters were optimized for lowest RRMSE with respect to the ground truth for each and every experiment in the simulation experiments and selected manually based on visual assessment in the neutron tomography experiment. The parameter expressing the level of trust (strength of smoothing) in the dynamic data (denoted by γ in [19]) was also manually tuned in the neutron tomography experiment. The rSIRT-PWC algorithm was implemented as described in Section III with the PWC estimation applied every 20 th iteration starting from the 60 th iteration.…”

“…A more statistical approach, i.e., a maximum a posteriori probability (MAP) estimation method, was presented in [17]. Other approaches do not rely on a prior high quality image, such as the spatialtemporal regularization approach based on non-local means proposed by Kazantsev et al [18], [19] and the region-based SIRT (rSIRT) method, which assumes only prior knowledge about the location of stationary voxels and dynamic voxels [20].…”