<div>There is an increasing need to accurately image objects at a high temporal resolution for different applications in order to analyze the underlying physical, chemical, or biological processes. In this thesis, we use advanced models exploiting the image structure and the measurement process in order to achieve an improved temporal resolution. The thesis is divided into three chapters, each corresponding to a different imaging application.</div><div><br></div><div>In the first chapter, we propose a novel method to localize neurons in fluorescence microscopy images. Accurate localization of neurons enables us to scan only the neuron locations instead of the full brain volume and thus improve the temporal resolution of neuron activity monitoring. We formulate the neuron localization problem as an inverse problem where we reconstruct an image that encodes the location of the neuron centers. The sparsity of the neuron centers serves as a prior model, while the forward model comprises of shape models estimated from training data.</div><div><br></div><div>In the second chapter, we introduce multi-slice fusion, a novel framework to incorporate advanced prior models for inverse problems spanning many dimensions such as 4D computed tomography (CT) reconstruction. State of the art 4D reconstruction methods use model based iterative reconstruction (MBIR), but it depends critically on the quality of the prior modeling. Incorporating deep convolutional neural networks (CNNs) in the 4D reconstruction problem is difficult due to computational difficulties and lack of high-dimensional training data. Multi-Slice Fusion integrates the tomographic forward model with multiple low dimensional CNN denoisers along different planes to produce a 4D regularized reconstruction. The improved regularization in multi-slice fusion allows each time-frame to be reconstructed from fewer measurements, resulting in an improved temporal resolution in the reconstruction. Experimental results on sparse-view and limited-angle CT data demonstrate that Multi-Slice Fusion can substantially improve the quality of reconstructions relative to traditional methods, while also being practical to implement and train.</div><div><br></div><div>In the final chapter, we introduce CodEx, a synergistic combination of coded acquisition and a non-convex Bayesian reconstruction for improving acquisition speed in computed tomography (CT). In an ideal ``step-and-shoot'' tomographic acquisition, the object is rotated to each desired angle, and the view is taken. However, step-and-shoot acquisition is slow and can waste photons, so in practice the object typically rotates continuously in time, leading to views that are blurry. This blur can then result in reconstructions with severe motion artifacts. CodEx works by encoding the acquisition with a known binary code that the reconstruction algorithm then inverts. The CodEx reconstruction method uses the alternating direction method of multipliers (ADMM) to split the inverse problem into iterative deblurring and reconstruction sub-problems, making reconstruction practical. CodEx allows for a fast data acquisition leading to a good temporal resolution in the reconstruction.</div>

Non-destructive imaging modalities for evaluating the internal properties of materials can be formulated as physics-driven inverse problems. Model-based Iterative reconstruction (MBIR) methods that integrate a forward model of the imaging system and a prior model of the object being imaged can provide superior reconstruction quality relative to conventional methods. However, making MBIR feasible for practical applications faces two key challenges. First, we require efficient computational methods for MBIR that allow large-scale reconstructions in real-time. Second, we must develop forward models that accurately capture the physics and geometry of the imaging system, and, support the use of advanced denoisers that enhance image quality as prior models.<br><br>This thesis attempts to address the aforementioned challenges and is divided into three main chapters, each corresponding to a different inverse imaging application. <br><br>In the first chapter of this thesis, we propose a novel 4D model-based iterative reconstruction (MBIR) algorithm for low-angle coherent-scatter X-ray Diffraction (XRD) tomography that can substantially increase the SNR. Our forward model is based on a Poisson photon counting model that incorporates a spatial point-spread function, detector energy response and energy-dependent attenuation correction. Our prior model uses a Markov random field (MRF) together with a reduced spectral bases set determined using non-negative matrix factorization. Our algorithm efficiently computes the Bayesian estimate by exploiting the sparsity of the measurement data. We demonstrate the ability of our method to achieve sufficient spatial resolution from sparse photon-starved measurements and also discriminate between materials of similar densities with real datasets.<br><br>In the second chapter of this thesis, we propose a multi-agent consensus equilibrium (MACE) algorithm for distributing both the computation and memory of <br>MBIR for Computed Tomographic (CT) reconstruction across a large number of parallel nodes. In MACE, each node stores only a sparse subset of views and a small portion of the system matrix, and each parallel node performs a local sparse-view reconstruction, which based on repeated feedback from other nodes, converges to the global optimum. Our distributed approach can also incorporate advanced denoisers as priors to enhance reconstruction quality. In this case, we obtain a parallel solution to the serial framework of Plug-n-play (PnP) priors, which we call MACE-PnP. In order to make MACE practical, we introduce a partial update method that eliminates nested iterations and prove that it converges to the same global solution. Finally, we validate our approach on a distributed memory system with real CT data. We also demonstrate an implementation of our approach on a massive supercomputer that can perform large-scale reconstruction in real-time. <br><br>In the third chapter of this thesis, we propose a method that makes MBIR feasible for real-time single-shot holographic imaging through deep turbulence. Our method uses surrogate optimization techniques to simplify and speedup the reflectance and phase-error updates in MBIR. Further, our method accelerates computation of the surrogate-updates by leveraging cache-prefetching and SIMD vector processing units on a single CPU core. We analyze the convergence and real CPU time of our method using simulated datasets, and demonstrate its dramatic speedup over the original MBIR approach. <br>

In this paper we deal with aspects of characterizing the ill-posedn ess of nonlinear inverse problems based on the discussion of specific examples. In particular, a parameter identification problem to a second order differential equation and its ill-posed linear components are under consideration. A new approach to the classification ofill-posedness degrees for multiplication operators completes the paper.

The stable solution of ill-posed non-linear operator equations in Banach space requires regularization. One important approach is based on Tikhonov regularization, in which case a one-parameter family of regularized solutions is obtained. It is crucial to choose the parameter appropriately. Here, a variant of the discrepancy principle is analyzed. In many cases such parameter choice exhibits the feature, called regularization property below, that the chosen parameter tends to zero as the noise tends to zero, but slower than the noise level. Here we shall show such regularization property under two natural assumptions. First, exact penalization must be excluded, and secondly, the discrepancy principle must stop after a finite number of iterations. We conclude this study with a discussion of some consequences for convergence rates obtained by the discrepancy principle under the validity of some kind of variational inequality, a recent tool for the analysis of inverse problems.

Title: Numerical Methods in Discrete Inverse Problems Author: Marie Kubínová Department: Department of Numerical Mathematics Supervisor: RNDr. Iveta Hnětynková, Ph.D., Department of Numerical Mathe- matics Abstract: Inverse problems represent a broad class of problems of reconstruct- ing unknown quantities from measured data. A common characteristic of these problems is high sensitivity of the solution to perturbations in the data. The aim of numerical methods is to approximate the solution in a computationally efficient way while suppressing the influence of inaccuracies in the data, referred to as noise, that are always present. Properties of noise and its behavior in reg- ularization methods play crucial role in the design and analysis of the methods. The thesis focuses on several aspects of solution of discrete inverse problems, in particular: on propagation of noise in iterative methods and its representation in the corresponding residuals, including the study of influence of finite-precision computation, on estimating the noise level, and on solving problems with data polluted with noise coming from various sources. Keywords: discrete inverse problems, iterative solvers, noise estimation, mixed noise, finite-precision arithmetic - iii -

Les problèmes de complétion de données interviennent dans divers domaines de la physique, tels que la mécanique, l'acoustique ou la thermique. La mesure directe des conditions aux limites se heurte souvent à l'impossibilité de placer l'instrumentation adéquate. La détermination de ces données n'est alors possible que grâce à des informations complémentaires. Des mesures surabondantes sur une partie accessible de la frontière mènent à la résolution d'un problème inverse de type Cauchy. Cependant, dans certains cas, des mesures directes sur la frontière sont irréalisables, des mesures de champs plus facilement accessibles permettent de pallier ce problème. Cette thèse présente des méthodes de régularisation évanescente qui permettent de trouver, parmi toutes les solutions de l'équation d'équilibre, la solution du problème de complétion de données qui s'approche au mieux des données de type Cauchy ou de champs partiels. Ces processus itératifs ne dépendent pas d'un coefficient de régularisation et sont robustes vis à vis du bruit sur les données, qui sont recalculées et de ce fait débruitées. Nous nous intéressons, dans un premier temps, à la résolution de problèmes de Cauchy associés à l'équation d'Helmholtz. Une étude numérique complète est menée, en utilisant la méthode des solutions fondamentales en tant que méthode numérique pour discrétiser l'espace des solutions de l'équation d'Helmholtz. Des reconstructions précises attestent de l'efficacité et de la robustesse de la méthode. Nous présentons, dans un second temps, la généralisation de la méthode de régularisation évanescente aux problèmes de complétion de données à partir de mesures de champs partielles. Des simulations numériques, pour l'opérateur de Lamé, dans le cadre des éléments finis et des solutions fondamentales, montrent la capacité de la méthode à compléter et débruiter des données partielles de champs de déplacements et à identifier les conditions aux limites en tout point de la frontière. Nous retrouvons des reconstructions précises et un débruitage efficace des données lorsque l'algorithme est appliqué à des mesures réelles issues de corrélation d'images numériques. Un éventuel changement de comportement du matériau est détecté grâce à l'analyse des résidus de déplacements. / Data completion problems occur in many engineering fields, such as mechanical, acoustical and thermal sciences. Direct measurement of boundary conditions is often confronting with the impossibility of placing the appropriate instrumentation. The determination of these data is then possible only through additional informations. Overprescribed measurements on an accessible part of the boundary lead to the resolution of an inverse Cauchy problem. However, in some cases, direct measurements on the boundary are inaccessible, to overcome this problem field measurements are more easily accessible. This thesis presents fading regularization methods that allow to find, among all the solutions of the equilibrium equation, the solution of the data completion problem which fits at best Cauchy or partial fields data. These iterative processesdo not depend on a regularization coefficient and are robust with respect to the noise on the data, which are recomputed and therefore denoised. We are interested initially in solving Cauchy problems associated with the Helmholtz equation. A complete numerical study is made, usingthe method of fundamental solutions as a numerical method for discretizing the space of the Helmholtz equation solutions. Accurate reconstructions attest to the efficiency and the robustness of the method. We present, in a second time, the generalization of the fading regularization method to the data completion problems from partial full-field measurements. Numerical simulations, for the Lamé operator, using the finite element method or the method of fundamental solutions, show the ability of the iterative process to complete and denoise partial displacements fields data and to identify the boundary conditions at any point. We find precise reconstructions and efficient denoising of the data when the algorithm is applied to real measurements from digital image correlation. A possible change in the material behavior is detected thanks to the analysis of the displacements residuals.

No description available.

Reconstruction of sonar images is an inverse problem, which is normally solved with model-based methods. These methods may introduce undesired artifacts called angular and range leakage into the reconstruction. In this thesis, a method called Learned Primal-Dual Reconstruction, which combines a data-driven and a model-based approach, is used to investigate the use of data-driven methods for reconstruction within sonar imaging. The method uses primal and dual variables inspired by classical optimization methods where parts are replaced by convolutional neural networks to iteratively find a solution to the reconstruction problem. The network is trained and validated with synthetic data on eight models with different architectures and training parameters. The models are evaluated on measurement data and the results are compared with those from a purely model-based method. Reconstructions performed on synthetic data, where a ground truth image is available, show that it is possible to achieve reconstructions with the data-driven method that have less leakage than reconstructions from the model-based method. For reconstructions performed on measurement data where no ground truth is available, some variants of the learned model achieve a good result with less leakage.

In this paper, we examine the feasibility of extending the Akaike information criterion (AIC) for deterministic systems as a potential model selection criteria for stochastic models. We discuss the implementation method for three different classes of stochastic models: continuous time Markov chains (CTMC), stochastic differential equations (SDE), and random differential equations (RDE). The effectiveness and limitations of implementing the AIC for comparison of stochastic models is demonstrated using simulated data from the three types of models and then applied to experimental longitudinal growth data for algae.

Thesis (Ph.D.)--Joint Program in Physical Oceanography (Massachusetts Institute of Technology, Dept. of Earth, Atmospheric, and Planetary Sciences and the Woods Hole Oceanographic Institution), 2000. / Includes bibliographical references (p. 179-187). / A simple dynamic model is used with various observations to provide an approximate spectral description of low frequency oceanic variability. Such a spectrum has wide application in oceanography, including the optimal design of observational strategy for the deployment of floats, the study of Lagrangian statistics and the estimate of uncertainty for heat content and mass flux. Analytic formulas for the frequency and wavenumber spectra of any physical variable, and for the cross spectra between any two different variables for each vertical mode of the simple dynamic model are derived. No heat transport exists in the model. No momentum flux exists either if the energy distribution is isotropic. It is found that all model spectra are related to each other through the frequency and wavenumber spectrum of the stream-function for each mode, ... , where ... represent horizontal wavenumbers, w stands for frequency, n is vertical mode number, and ... are latitude and longitude, respectively. Given ... , any model spectrum can be estimated. In this study, an inverse problem is faced: ... is unknown; however, some observational spectra are available. I want to estimate ... if it exists. Estimated spectra of the low frequency variability are derived from various measurements: (i) The vertical structure of and kinetic energy and potential energy is inferred from current meter and temperature mooring measurements, respectively. (ii) Satellite altimetry measurements produce the geographic distributions of surface kinetic energy magnitude and the frequency and wavenumber spectra of sea surface height. (iii) XBT measurements yield the temperature wavenumber spectra and their depth dependence. (v) Current meter and temperature mooring measurements provide the frequency spectra of horizontal velocities and temperature. It is found that a simple form for ... does exist and an analytical formula for a geographically varying ... is constructed. Only the energy magnitude depends on location. The wavenumber spectral shape, frequency spectral shape and vertical mode structure are universal. This study shows that motion within the large-scale low-frequency spectral band is primarily governed by quasigeostrophic dynamics and all observations can be simplified as a certain function of ... The low frequency variability is a broad-band process and Rossby waves are particular parts of it. Although they are an incomplete description of oceanic variability in the North Pacific, real oceanic motions with energy levels varying from about 10-40% of the total in each frequency band are indistinguishable from the simplest theoretical Rossby wave description. At higher latitudes, as the linear waves slow, they disappear altogether. Non-equatorial latitudes display some energy with frequencies too high for consistency with linear theory; this energy produces a positive bias if a lumped average westward phase speed is computed for all the motions present. / by Xiaoyun Zang. / Ph.D.