Low-dose spectral CT reconstruction based on image-gradient L₀-norm and adaptive spectral PICCS

Area of Science:

• Medical Imaging
• Computational Imaging
• Photon-Counting Detector Technology

Background:

• Spectral computed tomography (CT) using photon-counting detectors shows potential for advanced applications like lesion detection and material decomposition.
• Low signal-to-noise ratios in multi-energy projection data can degrade reconstructed image quality, limiting clinical utility.
• Existing spectral prior image constrained compressed sensing (SPICCS) methods using L1-norm regularization can lead to blurred edges.

Purpose of the Study:

• To develop an advanced low-dose spectral CT reconstruction algorithm that enhances image quality by preserving fine structures and edges.
• To address the limitations of existing SPICCS methods regarding edge blurring in reconstructed images.
• To introduce an adaptive weighting strategy to account for spectral differences across energy channels.

Main Methods:

• Incorporation of the image gradient L0-norm into the PICCS framework to improve edge preservation.
• Development of an adaptive weighting factor for channel-wise images to account for spectral energy differences.
• Implementation of the split-Bregman method for efficient objective function minimization.
• Evaluation using extensive numerical simulations and physical phantom experiments.

Main Results:

• The proposed L0-norm based adaptive SPICCS (L0-ASPICCS) algorithm demonstrated superior performance compared to simultaneous algebraic reconstruction technique, total variation minimization, and the original SPICCS.
• Qualitative and quantitative evaluations confirmed enhanced preservation of finer structures and sharper edges in reconstructed images.
• The adaptive weighting strategy effectively handled spectral differences, contributing to improved overall image quality.

Conclusions:

• The L0-ASPICCS algorithm represents a significant advancement in low-dose spectral CT reconstruction, offering improved image fidelity.
• This method effectively suppresses noise while preserving crucial image details, making it highly suitable for clinical applications.
• The adaptive L0-norm approach provides a robust solution for enhancing the diagnostic capabilities of spectral CT imaging.