An expectation maximization reconstruction algorithm for emission tomography with non-uniform entropy prior

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Emission tomography generates images from projection data, which is inherently noisy.
• Existing reconstruction algorithms like Maximum Likelihood (ML) have limitations in balancing global and local image quality.
• Maximum A Posteriori (MAP) approaches offer potential for improved image reconstruction but require careful implementation.

Purpose of the Study:

• To develop a Bayesian image reconstruction algorithm for emission tomography.
• To incorporate Poisson noise characteristics and non-uniform entropy priors into the reconstruction process.
• To achieve a compromise between the global performance of MAP and the local performance of ML algorithms.

Main Methods:

• A Bayesian image reconstruction algorithm using a Maximum A Posteriori (MAP) approach.
• Incorporation of Poisson noise in projection data and non-uniform entropy as a prior.
• Application of the Expectation Maximization (EM) method, solved using the Newton-Raphson numerical method.

Main Results:

• The proposed algorithm effectively incorporates Poisson noise and non-uniform entropy priors.
• Comparisons with ML demonstrated that the MAP algorithm provides a balance between global and local image reconstruction performance.
• The algorithm showed improved performance in narrow valley regions compared to standard ML methods over many iterations.

Conclusions:

• The developed Bayesian MAP algorithm offers a superior compromise for emission tomography image reconstruction.
• This method enhances image quality by effectively managing noise and prior information.
• The algorithm shows promise for applications requiring high fidelity in both global structure and local detail.