Filtered Backprojection Algorithm Can Outperform Iterative Maximum Likelihood Expectation-Maximization Algorithm.

Area of Science:

• Medical imaging
• Computational imaging
• Image reconstruction algorithms

Background:

• Iterative maximum-likelihood expectation-maximization (ML-EM) and filtered backprojection (FBP) are common image reconstruction algorithms.
• ML-EM typically yields superior image quality compared to standard FBP.
• The least-squared error (LSE) is a criterion for comparing reconstruction accuracy.

Purpose of the Study:

• To compare the performance of windowed FBP and ML-EM algorithms for image reconstruction.
• To determine optimal parameters for both algorithms under different noise conditions.
• To evaluate algorithm performance using the least-squared error criterion.

Main Methods:

• Computer simulations were employed to assess both algorithms.
• The best reconstruction for each algorithm was identified by optimizing parameters (ML-EM stopping iteration, windowed FBP parameters).
• Reconstructions were compared against the true image using the LSE criterion.

Main Results:

• For noisy Poisson projections, windowed FBP images were superior to ML-EM images.
• For noiseless projections, FBP algorithms outperformed ML-EM.
• Windowed FBP demonstrated the ability to achieve reconstructions closer to the true image than ML-EM.

Conclusions:

• Windowed FBP can outperform ML-EM in image reconstruction accuracy, particularly under noisy conditions.
• The selection of optimal parameters for windowed FBP and ML-EM when the true image is unknown is a significant open problem.
• Further research is needed to address parameter selection challenges for practical applications.