L0 constrained sparse reconstruction for multi-slice helical CT reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Bayesian Inference

Background:

• Multi-slice helical computed tomography (CT) is crucial for medical diagnostics.
• Reconstruction algorithms often require numerous projections, increasing scan time and radiation dose.
• Existing methods using L1 or L2 priors have limitations in handling sparse data and noise.

Purpose of the Study:

• To develop a novel Bayesian maximum a posteriori (MAP) method for multi-slice helical CT reconstruction.
• To leverage an L0-norm prior for improved reconstruction quality with a reduced number of projections.
• To enhance computational efficiency and convergence speed using surrogate potential functions.

Main Methods:

• Implementation of a Bayesian MAP framework for CT image reconstruction.
• Utilization of an L0-norm prior, approximated by surrogate potential functions, to promote sparsity.
• Application of a multi-slice helical CT acquisition model.
• Comparative analysis against L1-norm and L2-norm prior-based methods.

Main Results:

• The proposed L0-norm prior method yields high-quality reconstructions from highly sparse, noise-free projections.
• Significant improvements in reconstruction quality were observed even in the presence of noise compared to L1 and L2 priors.
• The use of surrogate functions accelerated convergence speed during the iterative reconstruction process.

Conclusions:

• The Bayesian MAP method with an L0-norm prior offers a robust solution for sparse-view CT reconstruction.
• This approach significantly enhances image quality, particularly in challenging low-projection scenarios.
• The method demonstrates superior performance over traditional L1 and L2 priors, especially under noisy conditions.