Noise analysis of MAP-EM algorithms for emission tomography

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Accurate modeling of photon noise propagation is vital for assessing image quality in Positron Emission Tomography (PET) and Single-Photon Emission Computed Tomography (SPECT).
• Previous work established theoretical models for the iterative Maximum Likelihood-Expectation Maximization (ML-EM) algorithm.

Purpose of the Study:

• To extend theoretical modeling of photon noise propagation to Maximum a Posteriori-Expectation Maximization (MAP-EM) algorithms.
• To analyze noise and bias propagation in specific MAP-EM implementations, including those with gamma and Gaussian priors.

Main Methods:

• Theoretical analysis of noise and covariance matrix propagation through iterative reconstruction algorithms.
• Application of linearizations to MAP-EM algorithms incorporating prior information.
• Validation using Monte Carlo simulations to compare theoretical estimates with sample estimates.

Main Results:

• Developed a theoretical framework for modeling photon noise propagation in MAP-EM algorithms.
• Demonstrated accurate theoretical predictions of mean and covariance matrix for MAP-EM with gamma and Gaussian priors.
• Validated the theoretical model against Monte Carlo simulations under practical noise and bias conditions.

Conclusions:

• The theoretical model effectively predicts the propagation of photon noise in MAP-EM iterative reconstruction.
• This approach is valuable for understanding and optimizing image quality in PET and SPECT imaging.
• The validated theory provides a robust tool for evaluating reconstruction algorithm performance.