Bayesian reconstruction of PET images: methodology and performance analysis

Area of Science:

• Medical Imaging
• Statistical Modeling
• Nuclear Medicine

Background:

• Accurate image reconstruction is crucial for Positron Emission Tomography (PET).
• Standard filtered backprojection (FBP) has limitations in quantitative accuracy.
• Bayesian methods offer a robust framework for complex imaging problems.

Purpose of the Study:

• To develop and evaluate a practical Bayesian statistical methodology for PET image reconstruction.
• To compare the quantitative performance of the Bayesian method against standard FBP.
• To demonstrate the application of the Bayesian method in clinical settings.

Main Methods:

• Image reconstruction using a Bayesian formulation with Poisson data and Markov random field priors.
• Pre-reconstruction calibration including attenuation correction, randoms estimation, and scatter computation.
• Reconstruction via a pre-conditioned conjugate gradient method.
• Quantitative analysis using a multi-compartment chest phantom and comparison with well counter data.

Main Results:

• The Bayesian protocol demonstrated substantial improvements in relative quantitation over FBP.
• Enhanced accuracy was particularly notable when using short transmission scans.
• The method was successfully applied to a clinical chest study.

Conclusions:

• The developed Bayesian methodology provides a practical and effective approach for PET image reconstruction.
• This method offers superior quantitative accuracy compared to conventional techniques.
• The Bayesian approach is particularly advantageous in scenarios with limited data acquisition time.