Data-Driven Gradient Regularization for Quasi-Newton Optimization in Iterative Grating Interferometry CT...

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• Grating interferometry CT (GI-CT) offers enhanced diagnostic information for breast cancer imaging beyond conventional absorption-based CT.
• Reconstructing GI-CT images is challenging due to ill-posed problems and high noise levels.
• Data-driven regularization combined with iterative reconstruction shows promise for medical imaging inverse problems.

Purpose of the Study:

• To develop and present a novel algorithm, GradReg, for improved image reconstruction in CT.
• To seamlessly integrate data-driven regularization with quasi-Newton solvers for enhanced performance.
• To apply and evaluate GradReg on both conventional breast CT and GI-CT.

Main Methods:

• Developed GradReg, an algorithm combining data-driven regularization with quasi-Newton solvers.
• Applied regularization in the gradient domain, allowing Hessian approximation on denoised data.
• Utilized quasi-Newton solvers for improved handling of ill-conditioned reconstruction problems.

Main Results:

• GradReg significantly improves dose efficiency for both conventional breast CT and GI-CT.
• The proposed method demonstrates superior performance in handling ill-posed inverse problems.
• GI-CT shows greater benefit from GradReg due to its sharper gradients and higher spatial frequency content.

Conclusions:

• GradReg offers a robust approach for enhancing image reconstruction in CT, particularly for dose efficiency.
• The gradient domain regularization strategy is advantageous when used with quasi-Newton solvers.
• GradReg is broadly applicable to various image reconstruction tasks relying on gradient-based updates.