Regularized image reconstruction algorithms for positron emission tomography

Area of Science:

• Medical Imaging
• Computational Science
• Image Reconstruction

Background:

• Positron emission tomography (PET) imaging requires accurate emission mean estimation.
• Existing algorithms like MLEM and penalized weighted least-squares have limitations in image quality and accuracy.

Purpose of the Study:

• To develop novel, regularized algorithms for improved emission mean estimation in PET.
• To enhance image contrast and reduce noise artifacts in PET reconstructions.

Main Methods:

• Developed two iterative algorithms: one minimizing a penalized maximum-likelihood (PML) objective function using surrogate functions, and another using an iteration-dependent penalty for smoothing.
• Implemented and compared algorithms against MLEM and penalized weighted least-squares using synthetic and phantom data.

Main Results:

• The proposed PML algorithm guarantees non-negative estimates and monotonic objective function decrease.
• The second algorithm effectively smoothed images while preserving important edges.
• Both proposed algorithms demonstrated superior contrast accuracy compared to MLEM and penalized weighted least-squares under fixed background noise levels.

Conclusions:

• The novel regularized algorithms offer significant improvements in image contrast accuracy for PET.
• These methods provide a more robust approach to emission mean estimation in the presence of noise.