Z-Index Parameterization for Volumetric CT Image Reconstruction via 3-D Dictionary Learning

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computational Imaging

Background:

• Low-dose X-ray cone-beam CT (CBCT) suffers from significant image noise, hindering diagnostic accuracy.
• Existing reconstruction methods struggle to suppress noise effectively while preserving structural details in low-dose CBCT.

Purpose of the Study:

• To develop and validate a novel 3-D dictionary learning (3-DDL) method for noise suppression in low-dose CBCT.
• To introduce an adaptive parameter selection strategy for optimizing the 3-DDL reconstruction process.
• To evaluate the potential for significant radiation dose reduction in CBCT imaging.

Main Methods:

• Incorporation of a sparse constraint based on a 3-D dictionary into a regularized iterative reconstruction framework.
• Development of a Z-curve analysis and Z-index parameterization (ZIP) method for adaptive regularization parameter selection.
• Comparative analysis of 3-D versus 2-D dictionary representation efficiencies and validation using real-world CBCT data.

Main Results:

• The 3-D dictionary demonstrated significantly higher representation efficiency compared to the 2-D dictionary, evidenced by narrower Laplacian distributions of coefficients.
• The proposed Z-index parameterization (ZIP) method successfully identified optimal regularization parameters via the Z-curve.
• The 3-DDL algorithm with ZIP achieved superior reconstruction quality, indicated by lower RMSE and higher SSIM, and enabled dose reduction up to 8x with improved contrast-noise ratio.

Conclusions:

• The proposed 3-DDL method with the ZIP strategy effectively suppresses noise and enhances image quality in low-dose CBCT.
• This technique offers a substantial reduction in radiation dose, potentially by a factor of 8, without compromising diagnostic information.
• The 3-DDL approach represents a significant advancement for safer and more effective CBCT imaging.