Virtual-Mask Informed Prior for Sparse-View Dual-Energy CT Reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Sparse-view sampling in dual-energy computed tomography (DECT) reduces radiation dose and increases speed but causes artifacts.
• Existing diffusion models for sparse-view reconstruction often lack global constraints, limiting image quality.

Purpose of the Study:

• To develop a novel dual-domain diffusion model for high-quality sparse-view DECT reconstruction.
• To address limitations of current image-domain diffusion models by incorporating global constraints.

Main Methods:

• Proposed a dual-domain virtual-mask informed diffusion model leveraging DECT's inter-channel correlation.
• Designed a virtual mask for perturbation operations on high- and low-energy data to create high-dimensional tensor priors.
• Implemented a dual-domain collaboration strategy integrating wavelet and projection domain information for structural and detail optimization.

Main Results:

• The proposed VIP-DECT method demonstrated excellent performance across multiple datasets.
• Under 30-view sparse sampling, VIP-DECT achieved at least a 1.02 dB improvement in Peak Signal-to-Noise Ratio (PSNR).
• The method enhanced Structural Similarity Index Measure (SSIM) by 1.91% under sparse sampling conditions.

Conclusions:

• The dual-domain virtual-mask informed diffusion model effectively enhances sparse-view DECT reconstruction quality.
• The integration of dual-domain information and virtual mask strategy overcomes limitations of existing methods.
• VIP-DECT offers a promising solution for artifact reduction and improved image fidelity in low-dose DECT applications.