This collaborative research project, conducted together with the groups of Prof. Heckel (TU Munich, Germany) and Prof. Schramm (KU Leuven, Belgium), aims to improve static and dynamic positron emission tomography (PET) image reconstruction. It addresses key challenges in PET imaging, such as high noise levels arising from limited acquisition times and dose constraints, by integrating machine learning (ML) with classical model-based iterative reconstruction techniques.

The project focuses on the development and validation of advanced ML approaches and variational reconstruction methods, striving for both innovation and robustness in static and dynamic PET imaging. A central element is the establishment of a high-quality, diverse, open PET raw data repository, complemented by open-source computational tools to support the global research community. This infrastructure will enable systematic evaluation of ML-based and traditional reconstruction methods and foster methodological progress and collaboration.

Furthermore, an international reconstruction benchmark challenge will be organized to promote the exploration and fair comparison of novel algorithms, with the goal of enhancing the diagnostic quality of PET. By combining data-driven techniques with classical methods and a detailed understanding of PET physics, this multidisciplinary project seeks to advance clinical applications of PET imaging for the benefit of patients and the broader scientific community.