Stage-by-Stage Wavelet Optimization Refinement Diffusion Model for Sparse-View CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Signal Processing

Background:

• Sparse-view CT reconstruction presents challenges due to limited data, often leading to image artifacts and reduced diagnostic accuracy.
• Existing diffusion models for CT reconstruction face training instability and local minima convergence issues when operating solely in the sinogram or image domains.

Purpose of the Study:

• To develop a more robust and stable diffusion model for sparse-view CT reconstruction.
• To leverage the benefits of wavelet transform for enhanced feature representation and model optimization.

Main Methods:

• The Stage-by-stage Wavelet Optimization Refinement Diffusion (SWORD) model was developed, integrating low- and high-frequency generative models within a unified mathematical framework.
• The model utilizes wavelet decomposition to process frequency components, ensuring stable training by avoiding direct operation on the sinogram.
• A three-stage optimization procedure (low-frequency generation, high-frequency refinement, domain transform) guides the reconstruction process.

Main Results:

• The SWORD model demonstrated superior performance in sparse-view CT reconstruction compared to existing state-of-the-art methods.
• Quantitative and qualitative experimental results confirmed the effectiveness and robustness of the proposed approach.
• The use of wavelet transform as a sparsity prior significantly improved the stability and convergence of the diffusion model.

Conclusions:

• The SWORD model offers an innovative and effective solution for sparse-view CT reconstruction, addressing the limitations of conventional diffusion models.
• Integrating wavelet transform within a staged optimization framework enhances model stability and reconstruction fidelity.
• The proposed method represents a significant advancement in the field of computational imaging for CT reconstruction.