Dissertation

Faster X-ray Computed Tomography in Real-World Dynamic Applications

This dissertation investigates how the efficiency of Computed Tomography (CT) can be improved for dynamic scientific and industrial applications.

Author

A.B.M. Graas

Date

04 February 2026

Links

Thesis in Leiden Repository

Examples of such applications are (i) rapidly moving processes, in which each instant in time effectively constitutes an individual reconstruction problem; (ii) iterative and learning-based algorithms, which generally demand more computational resources than conventional algorithms; and (iii) applications that employ CT repetitively, for example for detecting production defects on conveyor belts. This thesis comprises four chapters, each corresponding to a research project and publication. Chapter 2 discusses a set-up for high-speed tomography of fluidized beds. Chapter 3 presents an on-line deep learning method for real-time tomography. Chapter 4 introduces a new software package that enables tailored implementations of CT algorithms on graphics processing units (GPUs). Chapter 5 lastly pioneers a self-supervised deep learning method for noise removal of radiographs in scenarios where neither ground truths nor secondary noisy examples can be acquired.

• computed tomography
• dynamical systems
• fluidized beds
• gpu programming
• on-line learning
• real-time tomography
• self-supervised learning
• tomographic reconstruction
• x-ray imaging