Fast minimum variance estimator for limited angle CT image reconstruction.

M H Buonocore; W R Brody; A Macovski

doi:10.1118/1.594838

Home/Archive/PubMed

PubMed · 7290021

Fast minimum variance estimator for limited angle CT image reconstruction.

M H Buonocore·W R Brody·A Macovski

View original source ↗Export metadata ↓

Abstract

Many applications of diagnostic cross sectional imaging require that images be reconstructed from a limited number of projections (limited angle). Convolution back projection has been unsuitable in these applications. Methods for reconstruction based on stochastic estimation theory, such as the minimum variance estimator, use a discrete linear measurement model and are suitable for limited angle reconstruction. Unfortunately, the computational requirements of these methods have precluded their use. In this paper, starting from the general minimum variance estimator x = RxyRyy-1y, a computationally efficient (fast) estimator is derived for limited angle reconstruction by choosing Rxy and Ryy in the simplest way consistent with the geometric considerations of data acquisition. Minimum variance has in the past been precluded from use by the large amount of computation required to compute Ryy-1. With the fast estimator, the computation is avoided because Ryy has a particular form that allows factorization of the matrix into a product of matrices, each of which is easily inverted. A demonstration of the estimator for the reconstruction of sharp peaks is provided. Image quality is similar to that obtained with other methods.

Explore related subjects

Models, Theoretical Tomography, X-Ray Computed Journal Article

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

M H Buonocore, W R Brody, A Macovski. Fast minimum variance estimator for limited angle CT image reconstruction.. https://doi.org/10.1118/1.594838

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related citations

PubMed2007

Sensitivity of the solution of the Elder problem to density, velocity and numerical perturbations.

In this paper the Elder problem is studied with the purpose of evaluating the inherent instabilities associated with the numerical solution of this problem. Our focus is first on the question of the existence of a unique numerical solution for this problem, and second on the grid density and fluid density requirements necessary for a unique numerical solution. In particular we have investigated the instability issues associated with the numerical solution of the Elder problem from the following perspectives: (i) physical instability issues associated with density differences; (ii) sensitivity of the numerical solution to idealization irregularities; and, (iii) the importance of a precise velocity field calculation and the association of this process with the grid density levels that is necessary to solve the Elder problem accurately. In the study discussed here we have used a finite element Galerkin model we have developed for solving density-dependent flow and transport problems, which will be identified as TechFlow. In our study, the numerical results of Frolkovic and de Schepper [Frolkovic, P. and H. de Schepper, 2001. Numerical modeling of convection dominated transport coupled with density-driven flow in porous media, Adv. Water Resour., 24, 63-72.] were replicated using the grid density employed in their work. We were also successful in duplicating the same result with a less dense grid but with more computational effort based on a global velocity estimation process we have adopted. Our results indicate that the global velocity estimation approach recommended by Yeh [Yeh, G.-T., 1981. On the computation of Darcian velocity and mass balance in finite element modelling of groundwater flow, Water Resour. Res., 17(5), 1529-1534.] allows the use of less dense grids while obtaining the same accuracy that can be achieved with denser grids. We have also observed that the regularity of the elements in the discretization of the solution domain does make a difference in obtaining a unique stationary solution for this problem. The results of our study also indicate that the density differences are critical in the solution of the Elder problem and that high density differences lead to the physical instability that is inherent with this problem. Other than the physical instability associated with the level of density differences used in the Elder problem, the following two points should be considered in solving the Elder problem in a consistent manner: (i) strict attention should be paid to the vertical grid Peclet number in developing the criteria for convergent grid selection; and, (ii) with a globally continuous velocity calculation stable solutions can be obtained at lower grid densities.

Models, Theoretical↗

PubMed2007

Modular and dynamic functionalization of polymeric scaffolds.

The design and synthesis of multifunctionalized, architecturally controlled polymers is a prerequisite for a variety of future applications of polymeric materials. On the basis of Nature's use of self-assembly in the creation of biomaterials, this Account describes concepts that were developed over the past 5 years that utilize noncovalent interactions such as hydrogen bonding, ionic interactions, electrostatic interactions, metal coordination, and pi-pi stacking in modification of copolymer side-chains to obtain multifunctional polymeric materials, induce polymer morphology changes, and influence bulk-polymer properties.

Models, Theoretical↗

PubMed2006

Eigenvalue spectra of random matrices for neural networks.

The dynamics of neural networks is influenced strongly by the spectrum of eigenvalues of the matrix describing their synaptic connectivity. In large networks, elements of the synaptic connectivity matrix can be chosen randomly from appropriate distributions, making results from random matrix theory highly relevant. Unfortunately, classic results on the eigenvalue spectra of random matrices do not apply to synaptic connectivity matrices because of the constraint that individual neurons are either excitatory or inhibitory. Therefore, we compute eigenvalue spectra of large random matrices with excitatory and inhibitory columns drawn from distributions with different means and equal or different variances.

Models, Theoretical↗