Bayesian image reconstruction for emission tomography incorporating Good's roughness prior on massively parallel...

Area of Science:

• Medical Imaging
• Computational Science
• Algorithm Development

Background:

• The expectation-maximization (EM) algorithm is crucial for maximum-likelihood image reconstruction in emission tomography.
• Unconstrained EM suffers from long computation times and noise artifacts, limiting clinical use.

Purpose of the Study:

• To address computational demands and noise artifacts in EM-based emission tomography.
• To improve the efficiency and quality of maximum-likelihood image reconstruction.

Main Methods:

• Implemented the EM algorithm on single-instruction, multiple-data (SIMD) parallel architectures.
• Restructured superposition integrals as partial differential equations for efficient computation.
• Incorporated a Markov random field prior with Good's roughness penalty to mitigate noise.

Main Results:

• Demonstrated efficient EM algorithm implementation on SIMD parallel processors.
• Achieved reduced computation times compared to conventional and hypercube architectures.
• Successfully reduced noise artifacts in reconstructed images.

Conclusions:

• Massively parallel SIMD architectures offer a viable solution for computationally intensive EM image reconstruction.
• The combined approach of parallel processing and Markov random fields enhances emission tomography imaging.
• This optimized EM algorithm shows promise for routine clinical application.