Comparing implementations of penalized weighted least-squares sinogram restoration

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Computed Tomography (CT) scanners measure X-ray energy deposition, but data is prone to errors requiring preprocessing.
• Sinogram preprocessing aims to restore accurate line integrals from noisy CT measurements for improved image reconstruction.
• Penalized Poisson likelihood (PL) and penalized weighted least-squares (PWLS) are statistical restoration approaches for CT sinogram data.

Purpose of the Study:

• To develop and compare two computational methods for PWLS sinogram restoration in CT.
• To improve computational performance of PWLS restoration, particularly in standard-dose CT imaging.
• To evaluate PWLS sinogram restoration's effectiveness in correcting unmodeled artifacts and enhancing image quality.

Main Methods:

• Compared a direct matrix-inversion strategy using a closed-form PWLS solution.
• Investigated an iterative conjugate-gradient algorithm for PWLS sinogram restoration.
• Optimized both methods by exploiting matrix symmetry and sparseness specific to the sinogram-restoration problem.

Main Results:

• Both direct and iterative methods yielded similar image quality in terms of resolution-variance tradeoffs.
• Direct matrix inversion was computationally inefficient, requiring an order of magnitude more computation than iterative methods.
• The iterative conjugate-gradient approach demonstrated promising results, achieving computation times half that of previous PL implementations.

Conclusions:

• Iterative conjugate-gradient PWLS sinogram restoration, with optimized matrix operations, offers computational benefits.
• This iterative method is computationally superior to direct matrix PWLS inversion and PL restoration.
• PWLS sinogram restoration using conjugate gradients is a viable and efficient alternative for standard-dose CT imaging.