Robust residual-guided iterative reconstruction for sparse-view CT in small animal imaging

Area of Science:

• Medical Imaging
• Computational Imaging
• X-ray Computed Tomography

Background:

• Sparse-view computed tomography (CT) presents challenges in achieving high-fidelity image reconstruction due to limited projection data.
• Existing reconstruction algorithms often struggle with noise and artifacts when dealing with sparse-view datasets.
• Reducing scanning time and radiation dose in CT is crucial, especially for sensitive applications like small animal imaging.

Purpose of the Study:

• Introduce a novel, robust image reconstruction algorithm, RGIRT (residual-guided Golub-Kahan iterative reconstruction technique), specifically for sparse-view CT.
• Achieve high-fidelity image reconstruction from a limited number of projection views.
• Improve the efficiency and accuracy of CT image reconstruction in low-dose and limited-view scenarios.

Main Methods:

• RGIRT employs an inner-outer dual iteration framework.
• The inner iteration uses a flexible least squares QR (FLSQR) algorithm with Golub-Kahan bidiagonalization and weighted generalized cross-validation for hyper-parameter estimation.
• The outer iteration refines the solution by minimizing the residual using the intermediate reconstruction from the inner iteration.

Main Results:

• RGIRT demonstrated robust performance in reconstructing images from sparse-view CT data, outperforming reference methods like FBPConvNet, SART-TV, and FLSQR.
• Evaluations using numerical phantoms and experimental Micro-CT data showed consistent high-quality reconstructions across various projection view numbers and noise levels.
• Theoretical analysis confirmed the convergence of the residual, supporting the algorithm's stability and effectiveness.

Conclusions:

• RGIRT offers a significant advancement in sparse-view CT image reconstruction.
• The algorithm provides a robust solution for obtaining high-fidelity images with reduced projection data.
• This technique has the potential to shorten CT scanning times and lower radiation exposure in medical and research applications.