Dual-Domain Collaborative Diffusion Sampling for Multi-Source Stationary Computed Tomography Reconstruction

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Multi-source stationary CT offers high temporal resolution but faces sparse-view challenges due to limited source numbers.
• Existing sparse-view CT reconstruction methods often focus on either the sinogram or image domain, leading to limitations like artifacts or projection inaccuracies.

Purpose of the Study:

• To develop an advanced reconstruction method for sparse-view CT that overcomes the limitations of single-domain approaches.
• To introduce the Dual-domain Collaborative Diffusion Sampling (DCDS) model, integrating sinogram and image domain processing for improved sparse-view CT reconstruction.

Main Methods:

• The DCDS model employs a collaborative diffusion mechanism, enabling feedback between sinogram and image domains for enhanced reconstruction.
• It integrates sinogram recovery and image generative capabilities within an optimized mathematical framework.
• The model is optimized using the alternative direction iteration method for data consistency updates.

Main Results:

• The DCDS model demonstrated superior performance in reconstructing sparse-view CT data across numerical simulations, phantoms, and clinical cardiac datasets.
• It consistently outperformed state-of-the-art benchmarks in delivering high-quality reconstructions and accurate sinograms.
• The dual-domain approach effectively addressed artifacts and projection inaccuracies inherent in single-domain methods.

Conclusions:

• The DCDS model represents a significant advancement in sparse-view CT reconstruction by synergistically leveraging both sinogram and image domain information.
• This novel approach offers enhanced reconstruction quality and precision, paving the way for improved diagnostic capabilities in medical imaging.