Low-dose spectral CT reconstruction using image gradient ℓ ₀-norm and tensor dictionary

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Spectral computed tomography (CT) offers superior lesion detection and material decomposition.
• Existing tensor dictionary learning (TDL) methods struggle with preserving edge information in spectral CT reconstruction.

Purpose of the Study:

• To propose an improved tensor dictionary learning method for low-dose spectral CT reconstruction.
• To enhance edge preservation in spectral CT images.

Main Methods:

• Introduced a novel ℓ0TDL method incorporating a gradient ℓ0-norm constraint.
• Utilized the similarity of spectral CT images within the TDL framework.
• Employed the split-bregman method for optimization.

Main Results:

• The proposed ℓ0TDL method demonstrated superior performance compared to TV minimization, TV+LR, and TDL.
• Numerical simulations and real mouse studies validated the effectiveness of ℓ0TDL.
• The method successfully overcame the edge-preserving limitations of standard TDL.

Conclusions:

• The ℓ0TDL method significantly enhances low-dose spectral CT reconstruction quality.
• This approach offers improved lesion detection and tissue characterization capabilities.
• The ℓ0TDL method represents a promising advancement for clinical spectral CT applications.