The aim of the project is to improve the self-supervised reconstruction of computed tomography (CT) images under realistic imaging conditions. To this end, physical effects such as detector blur, spatially correlated image noise and a reduced number of projections are to be integrated directly into the reconstruction process. In addition, the study investigates whether joint modelling of detector correction and image reconstruction yields better results than conventional pre-processing steps.
The methods developed will be evaluated using the real-world 2DeteCT dataset. The long-term aim is to provide open-source tools that enable more robust CT image reconstruction at lower radiation doses, thereby contributing to safer and more reliable medical imaging.