Coded Aperture Collimation for High-Resolution Intraoperative Gamma Imaging

• Applicant: Dr.-Ing. Tobias Meißner
• Subject Area: Medical Physics, Biomedical Technology ; Methods in Artificial Intelligence and Machine Learning ; Nuclear Medicine, Radiotherapy, Radiobiology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 573323274

Gamma imaging is essential in nuclear medicine for visualizing metabolic processes using radiotracer that emit gamma rays. Intraoperative Gamma Cameras (IGCs) are designed to provide guiding imaging during surgery. The key challenge for IGCs is the need for high spatial resolution and high sensitivity, because they are employed in a photon-starved environment. This proposal aims at overcoming the sensitivity-resolution trade-off of traditional collimators by leveraging Coded Aperture Imaging (CAI). A coded aperture consists of a mask with multiple pinholes arranged in a complex pattern. It combines the high sensitivity of a larger aperture with the high spatial resolution of small apertures. Although image reconstruction is required, because CAI produces overlapping projections. However, non-linear effects (NLEs) emerge when imaging in close proximity to the camera or when imaging extended sources. NLEs lead to heavy artifacts that drastically decrease the image quality typically expressed with the Contrast-to-Noise Ratio (CNR). Current reconstruction methods either neglect NLEs or apply a simplified compensation technique. While the potential of Machine Learning (ML) approaches that implicitly consider NLEs has been demonstrated, a systematic investigation on extended sources has not yet been undertaken. Thus, my goal is to develop and evaluate robust reconstruction methods that effectively overcome NLE-induced artifacts enabling the high-resolution visualization of extended sources. The project is driven by a central hypothesis: When using a coded aperture collimator for the imaging of extended sources, incorporating NLEs into the reconstruction method substantially improves the CNR. To investigate this hypothesis, a coded aperture collimator will be designed and combined with a HEXITEC detector available through my host university (WP 1). Three novel reconstruction methods will then be developed (WP 2) by generating a system matrix that describes the image transformation on a pixel-to-pixel level. NLEs will be explicitly incorporated via mathematical models from my previously developed simulation framework. With this matrix, two ML-based approaches will be trained. One for removing the NLEs and another that performs a direct reconstruction. The methods will be systematically evaluated with highly-realistic simulations (WP 3) with the aim to explore the relationship between the CNR, source extensions, and number of detected photons. Finally, the device will be evaluated with real-world measurements of clinical phantoms. Confirmation of the hypothesis would have profound implications. It would demonstrate that the imaging of extended sources is not a fundamental limitation of CAI and that general source distributions can be captured. This would represent a significant advancement in the development of IGCs, where the reduced acquisition time and improved spatial resolution could enable a more precise tumor staging, for example.

• DFG Programme: WBP Fellowship
• International Connection: United Kingdom
• Host: Dr. Sarah Bugby