Abstract:

This project bridges data science and high-energy physics to tackle the challenge of detecting medium-energy gamma-rays and high-energy x-rays in the MeV range (0.1–100 MeV). The high stopping power required for MeV photon detection presents significant technological challenges, limiting current detector performance. This PhD project aims to develop an AI-driven electron tracking system that enhances radiation detection by integrating Physics-Informed Neural Networks (PINNs) with state-of-the-art 3D CdZnTe (CZT) drift strip detectors. The primary objective is to determine the recoil electron direction in Compton camera configurations, reducing uncertainty and significantly improving angular resolution. 
 
The research follows a three-stage methodology: (1) develop Geant4-based simulations to model electron tracks in radiation detectors, (2) design a semi-supervised PINN framework to train neural networks for near real-time electron trajectory reconstruction, and (3) validate the models through experimental data from DTU Space and the European Synchrotron Radiation Facility (ESRF). By replacing conventional offline algorithms with AI-enhanced real-time processing, this project improves measurement accuracy while reducing data storage, computational costs, and detector size. 
 
The impact of this research extends beyond space missions, as its advancements in high-resolution radiation detection have applications in medical imaging, particularly in Low Dose Molecular Breast Imaging (LDMBI) for early cancer detection, nuclear safety, and homeland security. By leveraging cutting-edge AI techniques and detector physics, this project will lay the groundwork to potentially revolutionize both space-based gamma-ray astronomy and next-generation medical imaging technologies.