Resolution enhancement of imaging small-scale portions in a compactly supported function.

The ambiguity involved in reconstructing an image from limited Fourier data is removed using a new technique that incorporates prior knowledge of the location of regions containing small-scale features of interest. The prior discrete Fourier transform (PDFT) method for image reconstruction incorporates prior knowledge of the support, and perhaps general shape, of the object function being reconstructed through the use of a weight function. The new approach extends the PDFT by allowing different weight functions to modulate the different spatial frequency components of the reconstructed image. The effectiveness of the new method is tested on one- and two-dimensional simulations.

Resolution enhancement in computerized tomographic imaging.

We consider the problem of reconstructing an object function f(r) from finitely many linear functional values. In our main application, the function f(r) is a tomographic image, and the data are integrals of f(r) along thin strips. Because the data are limited, resolution can be enhanced through the inclusion of prior knowledge. One way to do that, a generalization of the prior discrete Fourier transform (PDFT) method, was suggested in 1982 [SIAM J. Appl. Math.42,933 (1982)] but was found to be difficult to implement for the tomography problem, and that application was not pursued. Recent advances in approximating the PDFT make it possible to achieve the desired resolution enhancement in an easily implemented procedure.

Reconstruction of 2D PET data with Monte Carlo generated system matrix for generalized natural pixels.

In discrete detector PET, natural pixels are image basis functions calculated from the response of detector pairs. By using reconstruction with natural pixel basis functions, the discretization of the object into a predefined grid can be avoided. Here, we propose to use generalized natural pixel reconstruction. Using this approach, the basis functions are not the detector sensitivity functions as in the natural pixel case but uniform parallel strips. The backprojection of the strip coefficients results in the reconstructed image. This paper proposes an easy and efficient way to generate the matrix M directly by Monte Carlo simulation. Elements of the generalized natural pixel system matrix are formed by calculating the intersection of a parallel strip with the detector sensitivity function. These generalized natural pixels are easier to use than conventional natural pixels because the final step from solution to a square pixel representation is done by simple backprojection. Due to rotational symmetry in the PET scanner, the matrix M is block circulant and only the first blockrow needs to be stored. Data were generated using a fast Monte Carlo simulator using ray tracing. The proposed method was compared to a listmode MLEM algorithm, which used ray tracing for doing forward and backprojection. Comparison of the algorithms with different phantoms showed that an improved resolution can be obtained using generalized natural pixel reconstruction with accurate system modelling. In addition, it was noted that for the same resolution a lower noise level is present in this reconstruction. A numerical observer study showed the proposed method exhibited increased performance as compared to a standard listmode EM algorithm. In another study, more realistic data were generated using the GATE Monte Carlo simulator. For these data, a more uniform contrast recovery and a better contrast-to-noise performance were observed. It was observed that major improvements in contrast recovery were obtained with MLEM when the correct system matrix was used instead of simple ray tracing. The correct modelling was the major cause of improved contrast for the same background noise. Less important factors were the choice of the algorithm (MLEM performed better than ART) and the basis functions (generalized natural pixels gave better results than pixels).

An iterative transmission algorithm incorporating cross-talk correction for SPECT.

Simultaneous emission/transmission acquisitions in cardiac SPECT with a Tc99m/Gd153 source combination offer the capability for nonuniform attenuation correction. However, cross-talk of Tc99m photons downscattered into the Gd153 energy window contaminates the reconstructed transmission map used for attenuation correction. The estimated cross-talk contribution can be subtracted prior to transmission reconstruction or incorporated in the reconstruction algorithm itself. In this work, we propose an iterative transmission algorithm (MLTG-S) based on the maximum-likelihood gradient algorithm (MLTG) that explicitly accounts for this cross-talk estimate. Clinical images were acquired on a three-headed SPECT camera, acquiring Tc99m emission and Gd153 transmission images simultaneously. Subtracting the cross-talk estimate prior to transmission reconstruction can result in negative and zero values if the estimate is larger than or equal to the count in the transmission projection bin, especially with increased attenuator size or amount of cross-talk. This results in inaccurate attenuation coefficients for MLTG reconstructions with cross-talk subtraction. MLTG-S reconstructions on the other hand, yield better estimates of attenuation maps, by avoiding the subtraction of the cross-talk estimate. Comparison of emission slices corrected for nonuniform attenuation reveals that inaccuracies in the reconstructed attenuation map caused by cross-talk can artificially enhance the extra-cardiac activity, confounding the ability to visualize the left-ventricular walls.

The Influence of Photon Attenuation on Tumor-to-Background and Signal-to-Noise Ratios for SPECT Imaging.

Expanding on the work of Nuyts et. al [1], Bai et. al. [2], and Bai and Shao [3], who all studied the effects of attenuation and attenuation correction on tumor-to-background ratios and signal detection, we have derived a general expression for the tumor-to-background ratio (TBR) for SPECT attenuated data that have been reconstructed with a linear, non-iterative reconstruction operator O. A special case of this is when O represents discrete filtered back-projection (FBP). The TBR of the reconstructed, uncorrected attenuated data (TBR(no-AC)) can be written as a weighted sum of the TBR of the FBP-reconstructed unattenuated data (TBR(FBP)) and the TBR of the FBP-reconstructed "difference" projection data (TBR(diff)). We evaluated the expression for TBR(no-AC) for a variety of objects and attenuation conditions. The ideal observer signal-to-noise ratio (SNR(ideal)) was also computed in projection space, in order to obtain an upper bound on signal detectability for a signal-known-exactly/background-known-exactly (SKE/BKE) detection task. The results generally show that SNR(ideal) is lower for tumors located deeper within the attenuating medium and increases for tumors nearer the edge of the object. In addition, larger values for the uniform attenuation coefficient µ lead to lower values for SNR(ideal). The TBR for FBP-reconstructed, uncorrected attenuated data can both under- and over-estimate the true TBR, depending on several properties of the attenuating medium, including the shape of the attenuator, the uniformity of the attenuator, and the degree to which the data are attenuated.

Image reconstruction from limited Fourier data.

We consider the problem of reconstructing a function f with bounded support S from finitely many values of its Fourier transform F. Although f cannot be band limited since it has bounded support, it is typically the case that f can be modeled as the restriction to S of a sigma-band-limited function, say g. Our reconstruction method is based on such a model for f. Of particular interest is the effect of the choice of sigma > 0 on the resolution.

Image reconstruction: a unifying model for resolution enhancement and data extrapolation. Tutorial.

In reconstructing an object function F(r) from finitely many noisy linear-functional values integral of F(r)Gn(r)dr we face the problem that finite data, noisy or not, are insufficient to specify F(r) uniquely. Estimates based on the finite data may succeed in recovering broad features of F(r), but may fail to resolve important detail. Linear and nonlinear, model-based data extrapolation procedures can be used to improve resolution, but at the cost of sensitivity to noise. To estimate linear-functional values of F(r) that have not been measured from those that have been, we need to employ prior information about the object F(r), such as support information or, more generally, estimates of the overall profile of F(r). One way to do this is through minimum-weighted-norm (MWN) estimation, with the prior information used to determine the weights. The MWN approach extends the Gerchberg-Papoulis band-limited extrapolation method and is closely related to matched-filter linear detection, the approximation of the Wiener filter, and to iterative Shannon-entropy-maximization algorithms. Non-linear versions of the MWN method extend the noniterative, Burg, maximum-entropy spectral-estimation procedure.

Iterative image reconstruction using prior knowledge.

A method is proposed to reconstruct signals from incomplete data. The method, which can be interpreted both as a discrete implementation of the so-called prior discrete Fourier transform (PDFT) spectral estimation technique and as a variant of the algebraic reconstruction technique, allows one to incorporate prior information about the reconstructed signal to improve the resolution of the signal estimated. The context of diffraction tomography and image reconstruction from samples of the far-field scattering amplitude are used to explore the performance of the method. On the basis of numerical computations, the optimum choice of parameters is determined empirically by comparing image reconstructions of the noniterative PDFT algorithm and the proposed iterative scheme.