PhD Topics

Supervisors: Prof. Henry Chapman (DESY, UHH), Prof. Timo Gerkmann (UHH)

Advanced sources such as free-electron lasers produce intense and coherent beams of X-rays that are opening up new possibilities to image biological materials, such as single molecules, at atomic resolution. Since atomic-resolution lenses do not exist such methods usually rely upon retrieving the structural information encoded in the far-field coherent diffraction pattern

[Shechtman et al, 2015]. This intensity pattern corresponds to the Fourier magnitude of the object, and is thus an incomplete measurement since the spectral phase cannot be measured. To reconstruct the original structure, the missing phase information needs to be retrieved . A very similar problem of phase retrieval is encountered in speech source separation and enhancement, which has received increasing attention recently [Gerkmann et al, 2015]. Just in the last few years, modern machine learning methods have been successfully applied to the problem aiming to obtain faster and more accurate results.
The goal of this project is to explore common concepts in X-ray imaging and speech enhancement. Recent advances in the application of machine learning in speech analysis will be transferred and tailored to X-ray imaging, obtaining deeper insights into this challenging inverse problem and providing new opportunities for high-resolution structure determination of biological systems.

Supervisors: Prof. Dr. Martin Müller (HZG), Prof. Dr. Stephan Olbrich (UHH)

To support materials science applications, we will develop and integrate application-specific data stream and file interfaces as well as processing and visualization algorithms, and optimize them for scalability and usability, based on an existing in-situ parallel data streaming and extraction framework. It is planned to combine and tightly couple 3Dd image based data extraction methods with 3D geometry based analytics, in order to develop and integrate innovative hybrid methods, e. g. for segmentation purposes. These should reduce or avoid intermediate data and take advantage of complexity reduction (O(N3) volume/voxel → O(N2) surface/polygons), in order to efficiently and automatically generate abstractions of the characteristics of materials and its changes over time. To enable and support interdisciplinary usage and openness for scientific and public scenarios, efficient parallel compression and decompression will be integrated as part of the extraction of the resulting 3D geometric and/or symbolic representations, and a thin multi-platform client software will be provided for open access and remote 3D viewing.

Supervisors: Prof. Dr. Christian Schroer (DESY, UHH), Jun.-Prof. Dr. Mathias Trabs (UHH)

The goal of this PhD project is the development and application of denoising and deconvolution techniques to data obtained via different modes of lensless X-ray microscopy. The candidate will be involved in all stages of the data life cycle. This involves the acquisition at the experiment, reconstruction and evaluation of the data. The focus of this project is the improvement of the data reconstruction step. Currently, the reconstruction is carried out by iterative phasing methods which already perform well although the methodology ignores blurring effects and observation noise of the measurements. Taking into account these additional problems, deconvolution and denoising techniques in combination with the state-of-the-art phasing methods should improve the reconstruction of the original signal. For classical statistical (inverse) regression models, there are various deblurring and denoising procedures which are already well understood, including Fourier or spectral methods, non-parametric Bayes methods.

To adapt such methods in the more complex physical context, the present convolution operator has to be understood precisely which then allows for the construction, implementation and application of novel reconstruction methods e.g. statistical multiresolution estimation.

The candidate will integrate these new techniques in the existing workflow and evaluate their capabilities in terms of reconstruction enhancement and applicability to the constantly growing size of data sets. The success of these tests will depend also on the performance of the numerical implementation of the methods under survey. We are looking for a candidate who is interested in different modi of imaging and in applying mathematically and numerical challenging methods to data obtained by these methods.

Supervisors: Prof. Dr. Christian Schroer (DESY, UHH), Prof. Dr.-Ing. Tobias Knopp (TUHH)

The interpretation of most experimental data in X-ray microscopy requires solving an inverse problem, i. e., finding the physical properties of a sample from a more or less known model for the image formation. Solving such inverse problems is usually a challenging task since the underlying mathematical problem is ill-posed and in turn small perturbations in the measurement due to noise lead to large errors in the calculated solution of the inverse problem. To handle inverse problems, the underlying linear reconstruction problem has to be solved by applying prior knowledge. For instance it can be assumed that the solution is smooth or that it has minimal total variation. In the recent years, the classical solvers for inverse problems have been complemented by approaches that are based on machine learning. Based on large databases of typical images, the inverse imaging operator can be learned and efficiently evaluated. Within this project the goal is to implement, improve, and compare state-of-the art algorithms for the solution of the inverse problem in X-ray microscopy using real-world data. The algorithms will be benchmarked with respect to noise reduction and reconstruction speed. A major focus will be put investigation of artifacts that appear when the algorithms are applied to unusual but physically plausible data.

Supervisors: Prof. Dr. Michael Breuer (HSU), Dr. Philipp Neumann (HSU), Prof. Dr. Jochen Küpper  (CFEL, DESY, UHH), Dr. Muhamed Amin (CFEL, DESY)
 
Single-particle diffractive imaging (SPI) relies on high-density streams of individual aerosolized particles to collect sufficient numbers of diffraction patterns with short x-ray pulses. As the intense xray pulses destroy the intercepted particles, a constant, focused stream of particles is decisive for the recording of many high-quality diffraction patterns. Aerodynamic lens stacks (ALS), a series of orifices, are a widespread technique for obtaining collimated particle beams in SPI experiments, but optimization for certain parameter sets, e.g., particle sizes, flow rates, pressures, is difficult. The development of advanced aerosol injection techniques for better selectivity of the injection and particle control exploiting novel operation regimes, including very low temperatures is thus an important issue. Numerical simulations are an important tool to better understand and optimize ALS, guiding experimental optimization of the injection process. Due to the multiscale character and the wide range of flow states, e.g., continuum vs. free molecular regime, these calculations require advanced simulation methodologies and high-performance computing techniques. Multiscale and multiphysics simulations for a huge number of particles over a large phase space lead to a vast amount of data requiring efficient simulation as well as on-line and off-line data analysis. Simulations of actual experiments on the fluid-dynamic transport and manipulation of nanoparticles/biomolecules over wide ranges of temperatures (4–300 K) and particle sizes (3–300 nm) will utilize particle methods such as molecular dynamics and direct simulation Monte Carlo methods to solve the Boltzmann equation for finite Knudsen numbers Kn; new and more appropriate modeling assumptions will be derived. Further challenges include calculations of actual cooling rates/temperatures of particles in the cold-gas fields and additional simultaneous forces on the particles, e.g., external electric fields, requiring variable multiphysics approaches. 

Supervisors: Prof. Dr. Frank Steinicke (UHH), Dr. Wim Leemans (DESY, M – Division Leader), Dr. Reinhard Bacher (DESY) 

Modern particle accelerators or laser facilities are fairly complex research facilities producing a huge amount of operations and scientific data. While there is huge interest in developing data science methods to process this enormous amount of data, the underlying physical principles and the inherent complexity of the technical systems involved make the operation and maintenance of these facilities an enormous data science challenge itself. Today, operations are performed through traditional standard user interfaces (UI) providing limited performance for handling complex control and operation tasks.  In the proposed thesis, the candidate will explore novel forms of UIs to improve the operation of particle accelerators or laser facilities, in particular, by using mixed reality (MR) technologies such as virtual and augmented reality and three-dimensional (3D) spatial UIs. These novel MR-based UIs will be analyzed, designed implemented and evaluated in a human-centered design (HCD) process including domain experts and laboratory as well as field studies in particle accelerators and laser facilities. 

Supervisors: Prof. Dr. Martin Müller (HZG), Julian Moosmann (HZG), Prof. Dr.-Ing. Timo Gerkmann (UHH) 

Propagation-based phase contrast in near- and full-field X-ray imaging typically assumes the attenuation of X-rays transmitted by the sample to be negligible. However, most samples fulfill this condition only approximately due to residual absorption. Moreover, enhanced intensity contrast due to free-space propagation is beneficial for samples where absorption contrast is low or noisy or for samples which exhibit mixed contrast and where the soft tissue components require additional contrast enhancement such that their signal rises above the absorption signal of the surrounding material. Apart from certain approximations (small phase variations, small attenuation, small propagation distances, attenuation-phase duality), a direct algebraic inversion of the general problem as described by the Fresnel diffraction integral does not exist. For in situ or in vivo experiments, where deposited dose and/or scan time are critical, the acquisition of a second intensity measurement (at a different propagation distance or energy) is not feasible. For such experiments the inverse problem is ill-posed as phase and attenuation have to be reconstructed from a single measurement. The problem is related to phase retrieval in speech enhancement, where in the last few years machine learning techniques have been employed to considerably push the field forward. In this project, expertise in speech enhancement and X-ray tomography are combined to make improvements in both fields.