Thesis subjects

Studying physics & astronomy or engineering physics? Please contact Matthieu Boone () if you're interested in a master thesis with us. The subjects below are indicative and we encourage you to contact us for a personal conversation, in which we can discuss the possiblities and think of a subject best suited for you.

No thesis subject has an obligatory internship or mobility, but there are some possibilities for those that we can discuss.

Within the Radiation Physics group, we perform research on

  • Improvement of the micro/nano CT technique
  • Development of new types of CT scanners
  • Development of peripheral equipment for the CT scanners
  • Reconstruction algorithms
  • Phasecontrast imaging
  • 3D analysis of the scan results
  • Combination of CT with XRF to get chemical element information

More information on CT scanning

Iterative reconstruction of mixed phase and attenuation projection images in high resolution X-ray CT imaging

Promotor: Matthieu Boone

Contact:

Number of students: 1

At very high resolutions, X-ray refraction or phase contrast becomes visible along with the traditional contrast due to X-ray attenuation. Although this effect yields a clear visual contrast (by edge-enhancement), it gives rise to unrealistic reconstructions of the 3D volume, which is highly undesirable in quantitative analysis. For this reason, there has been a large interest in the international community in this phase contrast effect. However, the processing of the images and extraction of the refractive index is far from trivial, as the measured image is a mixed phase-and-amplitude image, containing both signals simultaneously.

The goal of this master thesis is to investigate the potential of iterative reconstruction algorithms for improving the reconstruction of both the phase and the amplitude signal. In these algorithms a simulation of the physical process of the image formation can be implemented to improve the quality of the reconstructed 3D volume. To achieve this, several approximate models are available, which can be used as initial input, or which can be combined. In order to be practically feasible in image reconstruction, the simulation of the effect needs to be sufficiently correct and fast. Apart from the reconstruction, this simulation approach can also be applied in an in-house developed framework for the simulation of CT scans.

Phase contrast image of a fly
Phase contrast image of a fly
Phase contrast image of a fly's leg
Phase contrast image of a fly's leg

Analysis of 4D tomographic datasets

Promotor: Matthieu Boone, Veerle Cnudde

Contact:

Number of students: 1

One of the micro-CT systems available at UGCT is a unique gantry-based system, where the sample remains stationary in its vertical position. This Environmental Micro-CT (EMCT) system is designed for fast and continuous acquisition of dynamic processes. Currently, every 12 seconds a new CT scan can be achieved, resulting in a true 4D dataset, where a 3D volume is visualized as a function of time. Some results of this imaging method can be found on the YouTube channel of the PProgRess research group. This system is currently mostly used in geological research, where fluid flow dynamics inside porous rock material is investigated. These processes are relevant to for example CO2 capture and oil and gas production. However, to obtain quantitative results from this massive amount of data, 3D analysis of the reconstructed volumes is required. Currently, each time frame is analyzed separately using an analysis script, yet this can be optimized by exploiting the time dimension.

The goal of this master thesis is to develop methods to exploit the time dimension in 4D CT image analysis and thus improve the quality of the results. This methodology can then be applied to both simulated data, where a ground truth is available, and real data of geological experiments.

An illustration of a dynamic CT dataset and ad hoc processing (from Bultreys et al., 2016)
An illustration of a dynamic CT dataset and ad hoc processing (from Bultreys et al., 2016)

Correction methods for beam hardening effects in micro-CT

Promotor: Matthieu Boone

Contact:

Number of students: 1

An X-ray source produces a polychromatic beam, and the attenuation inside a sample is strongly energy-dependent. To this date however, most X-ray detectors used in tomography can't discriminate between X-rays of different energies, and for this and other reasons most reconstruction algorithms are based on the false assumption of ideal exponentional attenuation of a monochromatic beam. This causes so-called beam-hardening artefacts wich manifest themselves in an underestimation of the attenuation in the centre of the sample, and streaking along edges and between dense inclusions. Although there exist energy-dispersive X-ray detectors that could in theory resolve the different energies, they are still far away from practical use in high-resolution CT.

The purpose of this master thesis is therefore to evaluate, compare and improve existing beam-hardening correction methods using simulations of CT-scans and experimental data. A realistic CT projection simulator has been developed at UGCT which takes into account the polychromatic nature of the generated X-ray beams, energy-dependent attenuation in the sample as well as the spectral sensitivity of the detector The data it generates allows to test beam hardening correction methods since in a simulation one knows the "ground truth". There is also an alternative beam hardening correction method available at UGCT based on Fourier analysis of the reconstruction at two extreme settings which needs to be further evaluated in terms of accuracy, practicality and flexibility. Experimental data, possibly from energy-dispersive detectors, could be used to validate the simulated projections, or vice-versa, the simulation could offer valuable input parameters for existing beam-hardening methods.

A reconstructed slice of a dental implant without (left) and with (right) beam hardening correction applied.
A reconstructed slice of a dental implant without (left) and with (right) beam hardening correction applied.

Study on the use of prior knowledge about the object in iterative tomographic reconstruction algorithms

Promotor: Matthieu Boone, Veerle Cnudde

Contact:

Number of students: 1

Recently, iterative reconstruction algorithms for the processing of raw projection data into a 3D volume have gained interest. In these algorithms, projections of a temporary reconstructed volume are simulated and compared with the measured data. The difference between these two is used as input to update the temporary volume, until convergence is reached. In the simulation step, physical processes can be implemented to improve the similarity of the simulation and the real imaging process, hence to improve the quality of the reconstruction. For standard micro-CT scans, an empty volume is used as initial solution in order to avoid biasing in the reconstruction. However, when only limited data is available from the micro-CT scan, it can prove beneficial to implement prior knowledge about the sample, e.g. parts of the morphology.

This methodology is associated with the processing of dynamic CT scanning in which a (relatively) fast dynamic process is imaged. Some results of this imaging method can be found on the YouTube channel of the PProgRess research group: https://www.youtube.com/channel/UCUZ8aQ7pxuXuCIbPPTsngWQ. Typically, a high quality micro-CT scan is acquired before the process is initiated, followed by fast scans during the dynamic process. The latter scans are often low quality due to motion artefacts and limited photon statistics. However, a large part of the reconstructed volume remains unaltered with respect to the initial structure, hence this information can be used to improve the reconstruction quality of the fast CT scans.

The goal of this master thesis is to use prior knowledge about the initial structure of an object in dynamic CT scanning in the framework of fluid-flow experiments in geomaterials. In these experiments, multiple fluids (e.g. oil and water) are pumped through the pore space of a porous rock and fluid dynamics are investigated. In this process, the stone matrix remains constant and changes only occur within the pore space. During this work, the influence of using this prior knowledge on the reconstruction quality will be investigated. This will in the first place be investigated using phantom data to avoid effects such as matrix displacement due to the pressure buildup for the pumping and motion artefacts. The final goal is to improve the reliability of the segmentation of different fluids in the pore space on real data.

A video of the advection and diffusion of salt in a limestone

 

Reconstruction and analysis of hyperspectral datasets

Promotor: Matthieu Boone

Contact:

Number of students: 1

In typical X-ray detector systems, the X-ray energy of all incident photons is integrated, hence all spectral information is lost. The RP group has access to a novel energy-dispersive full-field detector, of which only a few systems exist worldwide. This detector records a series of 1000 2D images at different energies simultaneously, at an extraordinary energy resolution. However, this huge amount of data needs to be processed properly in order to exploit its full potential, which contains spectral hence chemical information about the object.

The goal of this master thesis is to develop methods to exploit the spectral information contained in these datasets. To achieve this, the complete spectral dataset, which contains 3 spatial dimensions and one spectral dimension, needs to be considered as a whole. These methods can be developed based on simulated data, but real datasets are available for real tests. Additional tests can also be performed during the course of this master thesis.

A single radiographic hyperspectral image, which is in fact a 3D datacube with the X-ray energy as the 3th dimension
A single radiographic hyperspectral image, which is in fact a 3D datacube with the X-ray energy as the 3th dimension
A rendering of an aortic arch in which gold nano-particles are visualized in subfigure c
A rendering of an aortic arch in which gold nano-particles are visualized in subfigure c

 

Calibration of the scan images in Hounsfield Units (HU)

Promotor: Matthieu Boone

Contact:

Number of students: 1

The result of a CT scan is a virtual 3D representation of the inside of the sample, composed of voxels (three dimensional pixels). Each voxel contains a calculated grey value which represents the local linear attenuation coefficient for the materials in that voxel. Medical CT scan images are expressed in Hounsfield Units (HU), which are based on a linear transformation of the attenuation coefficients of water and air. However, due to the differences in measurement conditions, the HU values of denser material such as bone is mapped to a different range in micro-CT as it is in medical CT. Furthermore, the HU values will depend on the acquisition settings. Therefore, comparing data is only possible with visual inspection, and a quantitative comparison between medical scans and scans performed at UGCT is impossible.

The goal of this master thesis is to develop methods to convert the reconstructed files from attenuation coefficients to Hounsfield Units by creating lookup-tables based on simulated data, which can afterwards be compared with experimental data. To achieve this, challenges such as beam hardening and the corresponding post-processing will need to be tackled as well.

Internship

The possiblities for internships are

  • Controlling hardware components in cooperation with the company X-Ray Engineering
  • Beamtime at a X-ray fluorescence beamline at a European synchrotron