Enhanced 3D PET OSEM reconstruction using inter-update Metz filtering

Area of Science:

• Medical Imaging
• Computational Physics
• Algorithm Development

Background:

• Ordered Set Expectation Maximization (OSEM) is a standard algorithm for 3D Positron Emission Tomography (PET) reconstruction.
• Improving the contrast-noise balance in PET images is crucial for accurate diagnosis and analysis.
• Existing OSEM algorithms may require post-reconstruction filtering, which can introduce artifacts or blur details.

Purpose of the Study:

• To introduce an enhanced OSEM algorithm, termed Inter-Update Metz Filtered OSEM (IMF-OSEM), for 3D PET reconstruction.
• To improve image quality by incorporating filtering directly into the iterative reconstruction process.
• To develop and evaluate a high-speed software implementation of the IMF-OSEM algorithm.

Main Methods:

• Developed the IMF-OSEM algorithm, which applies Metz filtering to image estimates at intervals during the OSEM iterative updates.
• Implemented a software package featuring optimized forward/back projection using symmetry and incremental computation.
• Utilized parallel processing capabilities for significant reconstruction acceleration, largely independent of data size.

Main Results:

• IMF-OSEM demonstrated improved contrast-noise balance in specific regions of interest compared to standard OSEM and post-filtered OSEM.
• The software implementation achieved reasonable reconstruction times, even with large datasets and non-axially compressed data.
• Performance was validated using phantom data from a GE Advance PET scanner.

Conclusions:

• The IMF-OSEM algorithm offers a superior approach to 3D PET image reconstruction by integrating filtering within the iterative process.
• Appropriate selection of Metz filter parameters is key to optimizing the contrast-noise trade-off.
• The developed software provides an efficient and accelerated solution for high-quality 3D PET reconstruction.