Weighted least-squares reconstruction methods for positron emission tomography

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Positron Emission Tomography (PET) imaging relies on accurate image reconstruction algorithms.
• Existing methods like Maximum Likelihood Expectation-Maximization (ML-EM) have limitations in speed and image quality.
• Development of advanced reconstruction techniques is crucial for improving diagnostic capabilities in PET.

Purpose of the Study:

• To introduce novel unpenalized and penalized weighted least-squares (WLS) reconstruction methods for PET.
• To incorporate parameter-dependent covariance-based weights and a priori information into PET reconstruction.
• To evaluate the performance of these WLS methods against established algorithms like ML-EM.

Main Methods:

• Developed unpenalized and penalized Weighted Least-Squares (WLS) objective functions for PET image reconstruction.
• Weights in WLS are derived from the covariance of model error, dependent on unknown parameters.
• Penalty functions incorporate a priori information for the penalized WLS method.
• Algorithms guarantee nonnegative estimates and monotonic decrease of the objective function for the unpenalized method.

Main Results:

• Experimental results show WLS methods converge faster than the ML-EM algorithm.
• Reconstructed PET images exhibit significantly improved resolution and contrast using WLS methods.
• Simulations suggest global convergence, though a formal proof is pending.
• The unpenalized WLS method demonstrates monotonic decrease in the objective function with iterations.

Conclusions:

• The proposed unpenalized and penalized WLS methods represent a significant advancement in PET image reconstruction.
• These WLS techniques offer superior performance in terms of speed, resolution, and contrast compared to ML-EM.
• Further investigation into the convergence properties of the proposed algorithms is warranted.