Dominique Van de Sompel

Task-based performance analysis of FBP, SART and ML for digital breast tomosynthesis using signal CNR and Channelised Hotelling Observers

We assess the performance of filtered backprojection (FBP), the simultaneous algebraic reconstruction technique (SART) and the maximum likelihood (ML) algorithm for digital breast tomosynthesis (DBT) under variations in key imaging parameters, including the number of iterations, number of projections, angular range, initial guess, and radiation dose. This is the first study to compare these algorithms for the application of DBT. We present a methodology for the evaluation of DBT reconstructions, and use it to conduct preliminary experiments investigating trade-offs between the selected imaging parameters. This investigation includes trade-offs not previously considered in the DBT literature, such as the use of a stationary detector versus a C-arm imaging geometry. A real breast CT volume serves as a ground truth digital phantom from which to simulate X-ray projections under the various acquisition parameters. The reconstructed image quality is measured using task-based metrics, namely signal CNR and the AUC of a Channelised Hotelling Observer with Laguerre-Gauss basis functions. The task at hand is the detection of a simulated mass inserted into the breast CT volume. We find that the image quality in limited view tomography is highly dependent on the particular acquisition and reconstruction parameters used. In particular, we draw the following conclusions. First, we find that optimising the FBP filter design and SART relaxation parameter yields significant improvements in reconstruction quality from the same projection data. Second, we show that the convergence rate of the maximum likelihood algorithm, optimised with paraboloidal surrogates and conjugate gradient ascent (ML-PSCG), can be greatly accelerated using view-by-view updates. Third, we find that the optimal initial guess is algorithm dependent. In particular, we obtained best results with a zero initial guess for SART, and an FBP initial guess for ML-PSCG. Fourth, when the exposure per view is constant, increasing the total number of views within a given angular range improves the reconstruction quality, albeit with diminishing returns. When the total dose of all views combined is constant, there is a trade-off between increased sampling using a larger number of views and increased levels of quantum noise in each view. Fifth, we do not observe significant differences when testing various access ordering schemes, presumably due to the limited angular range of DBT. Sixth, we find that adjusting the z-resolution of the reconstruction can improve image quality, but that this resolution is best adjusted by using post-reconstruction binning, rather than by declaring lower-resolution voxels. Seventh, we find that the C-arm configuration yields higher image quality than a stationary detector geometry, the difference being most outspoken for the FBP algorithm. Lastly, we find that not all prototype systems found in the literature are currently being run under the best possible system or algorithm configurations. In other words, the present study demonstrates the critical importance (and reward) of using optimisation methodologies such as the one presented here to maximise the DBT reconstruction quality from a single scan of the patient.

Modelling of experimentally created partial-thickness human skin burns and subsequent therapeutic cooling: a new measure for cooling effectiveness.

Rapid post-injury cooling of a skin burn has been shown to have both symptomatic and therapeutic benefits. However, the latter cannot be explained by temperature reduction alone, and must thus be secondary to an altered biological response. In this study, we construct a computational model to calculate the heat transfer and damage accumulation in human skin during and after a burn. This enables us to assess the effectiveness of various cooling protocols (involving both free and forced convection to air and water respectively) in terms of their reduction in Arrhenius tissue damage. In this process, we propose an extension of the Arrhenius damage model in the form of a new measure xi, which estimates the relevance of post-burn accrued damage. It was found that the reduction in Arrhenius damage integrals near the skin surface was too small to be physiologically relevant. Hence our results confirm that while the reduction in tissue temperatures is indeed quicker, the therapeutic benefit of cooling cannot be explained by thermal arguments (i.e. based on Arrhenius damage models) alone. We plan to validate this hypothesis by conducting future microarray analyses of differential gene expression in cooled and non-cooled burn lesions. Our computational model will support such experiments by calculating the necessary conditions to produce a burn of specified severity for a given experimental setup.

Regularising limited view tomography using anatomical reference images and information theoretic similarity metrics

This paper is concerned with limited view tomography. Inspired by the application of digital breast tomosynthesis (DBT), which is but one of an increasing number of applications of limited view tomography, we concentrate primarily on cases where the angular range is restricted to a narrow wedge of approximately ±30°, and the number of views is restricted to 10-30. The main challenge posed by these conditions is undersampling, also known as the null space problem. As a consequence of the Fourier Slice Theorem, a limited angular range leaves large swathes of the objects Fourier space unsampled, leaving a large space of possible solutions, reconstructed volumes, for a given set of inputs. We explore the feasibility of using same- or different-modality images as anatomical priors to constrain the null space, hence the solution. To allow for different-modality priors, we choose information theoretic measures to quantify the similarity between reconstructions and their priors. We demonstrate the limitations of two popular choices, namely mutual information and joint entropy, and propose robust alternatives that overcome their limitations. One of these alternatives is essentially a joint mixture model of the image and its prior. Promising mitigation of the data insufficiency problem is demonstrated using 2D synthetic as well as clinical phantoms. This work initially assumes a priori registered priors, and is then extended to allow for the registration to be performed simultaneously with the reconstruction.

A systematic performance analysis of the simultaneous algebraic reconstruction technique (SART) for limited angle tomography

The design of limited angle tomography systems requires the optimization of various imaging parameters in order to achieve useful as well as reliable results. Algebraic reconstruction techniques, specifically the SART algorithm, have given excellent results in CT and are being actively considered for limited angle commercial applications such as tomosynthesis. In this study, we simulate a range of limited angle scenarios by systematically varying a number of key imaging parameters, and examine the performance of the SART algorithm under these variations. The phantoms used are basic ellipsoids in 2D, yielding analytical projections, and an MR-derived breast phantom in 3D.

Robust Joint Entropy Regularization of Limited View Transmission Tomography Using Gaussian Approximations to the Joint Histogram

Information theoretic measures to incorporate anatomical priors have been explored in the field of emission tomography, but not in transmission tomography. In this work, we apply the joint entropy prior to the case of limited angle transmission tomography. Due to the data insufficiency problem, the joint entropy prior is found to be very sensitive to local optima. Two methods for robust joint entropy minimization are proposed. The first approximates the joint probability density function by a single 2D Gaussian, and is found to be appropriate for reconstructions where the ground truth joint histogram is dominated by two clusters, or multiple clusters that are roughly aligned. The second method is an extension to the case of multiple Gaussians. The intended application for the single Gaussian approximation is digital breast tomosynthesis, where reconstructed volumes are approximately bimodal, consisting mainly of fatty and fibroglandular tissues.

Systematic Performance Analysis of SART as Applied to Digital Breast Tomosynthesis

Breast tomosynthesis reconstructions require a wide range of design choices to be made in order to yield consistently good results. Since algebraic reconstruction, specifically the SART algorithm, has given excellent results in CT and is being actively considered for commercial use in tomosynthesis, we in this study systematically vary a number of key imaging parameters and assess the performance of the SART algorithm under these variations. The phantoms used are basic ellipsoids in 2D (yielding analytical projections), and an MR-derived breast phantom in 3D.

Simultaneous reconstruction and registration algorithm for limited view transmission tomography using a multiple cluster approximation to the joint histogram with an anatomical prior

We develop a novel simultaneous reconstruction and registration algorithm for limited view transmission tomography. We derive a cost function using Bayesian probability theory, and propose a similarity metric based on the explicit modeling of the joint histogram as a sum of bivariate clusters. The resulting algorithm shows a robust mitigation of the data insufficiency problem in limited view tomography. To our knowledge, our work represents the first attempt to incorporate non-registered, multimodal anatomical priors into limited view transmission tomography by using joint histogram based similarity measures.

Task-based performance analysis of SART for digital breast tomosynthesis using signal CNR and channelised Hotelling observers

In this study, we examine the performance of the simultaneous algebraic reconstruction technique (SART) for digital breast tomosynthesis under variations in key imaging parameters, such as the number of iterations, number of projections, angular range, initial guess, radiation dose, etc. We use a real breast CT volume as a ground truth digital phantom from which to simulate x-ray projections under the various selected conditions. The reconstructed image quality is measured using task-based metrics, namely signal CNR and the AUC of a Channelised Hotelling Observer with Laguerre-Gauss basis functions. The task at hand is a signal-known-exactly (SKE) task, where the objective is to detect a simulated mass inserted into the breast CT volume.

Avoiding local optima of the joint entropy prior in limited view tomography using a multiresolution scheme

The incorporation of anatomical reference images into limited view transmission tomography has been attempted previously by using the joint entropy prior. However, this prior has been found to be sensitive to local optima. Here, we propose to increase robustness to local optima by using a multiresolution optimisation scheme. To our knowledge, this is the first work to apply multiresolution optimisation to the joint entropy prior in limited view transmission tomography. The results show a substantial mitigation of the sensitivity to local optima, as well as a robustness to missing as well as extra regions in the anatomical reference image. In addition, we demonstrate the methods robustness to misalignment between the reconstruction and the anatomical reference image.

Robust incorporation of anatomical priors into limited view tomography using multiple cluster modelling of the joint histogram

We apply the joint entropy prior to limited view transmission tomography and demonstrate its sensitivity to local optima. We propose to increase robustness by modelling the joint histogram as the sum of a limited number of bivariate clusters. The method is illustrated for the case of Gaussian distributions. This approximation increases robustness by reducing the possible number of local optima in the cost function. The resulting reconstruction prior mimicks the behaviour of the joint entropy prior in that it narrows clusters in the joint histogram, and yields promisingly accurate reconstruction results despite the null space problem.