Resolving inter-crystal scatter in a light-sharing depth-encoding PET detector.

Area of Science:

• Medical Imaging
• Nuclear Physics
• Computer Science

Background:

• Inter-crystal scattering (ICS) in light-sharing PET detectors causes positioning ambiguity, degrading image quality.
• Existing methods struggle to accurately pinpoint initial interactions, limiting PET performance.

Purpose of the Study:

• To resolve ICS positioning ambiguity in Prism-PET's depth-encoding detector.
• To enhance contrast, quantitative accuracy, and spatial resolution in PET imaging.

Main Methods:

• Developed an analytical Bayesian estimation framework and a neural network for ICS positioning.
• Generalized to 3D event positioning using a convolutional neural network (CNN) for crystal identification and depth-of-interaction (DOI).

Main Results:

• Bayesian method yielded 20.5 keV energy and 3.1 mm DOI error.
• CNN achieved 83% crystal identification accuracy and 3.0 mm DOI error.
• Demonstrated 38.9% improvement in spatial resolution and enhanced hot spot clarity.

Conclusions:

• The Prism-PET's segmented light guide enables confined light-sharing, crucial for ICS recovery.
• The detector response model is key to achieving high accuracy in ICS positioning.
• This algorithm significantly improves PET image quality beyond previous methods.