A generalized reconstruction framework for unconventional PET systems.

Area of Science:

• Medical Imaging
• Nuclear Physics
• Computer Science

Background:

• Quantitative positron emission tomography (PET) imaging relies on accurate physical modeling.
• Designing novel PET geometries is crucial for specialized applications.
• Existing reconstruction methods often lack flexibility for arbitrary system designs.

Purpose of the Study:

• To develop a generalized reconstruction framework for PET systems with arbitrary detector geometries.
• To enable precise modeling of PET physics for unconventional systems.
• To facilitate the design and validation of novel PET scanner configurations.

Main Methods:

• Developed a generalized reconstruction framework using a Cartesian grid for image voxels.
• Implemented maximum likelihood-expectation-maximization (MLEM) algorithm for image reconstruction.
• Utilized Siddon's algorithm and a symmetry-seeking approach for efficient system matrix computation and memory optimization.
• Parallelized computations using open multiprocessing and message passing interface (MPI).

Main Results:

• Demonstrated the framework's robustness on three novel PET systems: a virtual-pinhole half-ring insert, a virtual-pinhole flat-panel insert, and a plant imaging PET system.
• Achieved significant memory and storage compression (up to 50x) using symmetry-seeking algorithms.
• Showcased the ability to model diverse and unconventional PET geometries.

Conclusions:

• The generalized framework advances arbitrary geometry reconstruction in PET.
• It reduces the effort needed to investigate and validate new PET system designs.
• Optimizations enhance computational efficiency (memory usage and speed) for novel PET scanners.