Mitigating Data Consistency Induced Discrepancy in Cascaded Diffusion Models for Sparse-View CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Healthcare

Background:

• Sparse-view Computed Tomography (CT) reduces radiation dose but causes image degradation.
• Diffusion models offer potential for CT reconstruction but are computationally intensive and face training-sampling discrepancies.

Purpose of the Study:

• To introduce a novel Cascaded Diffusion with Discrepancy Mitigation (CDDM) framework for high-quality sparse-view CT image reconstruction.
• To address computational costs and training-sampling discrepancies inherent in diffusion model-based CT reconstruction.

Main Methods:

• Developed a CDDM framework with cascaded low-quality (latent space) and high-quality (pixel space) image generation.
• Incorporated data consistency and discrepancy mitigation in a one-step reconstruction process.
• Utilized Alternating Direction Method of Multipliers (ADMM) for targeted regularization of image gradients.

Main Results:

• CDDM framework minimizes computational costs by inferring in latent space.
• Discrepancy mitigation ensures data distribution aligns with the original diffusion manifold.
• Experimental results show superior high-quality image generation with clearer boundaries compared to existing methods.

Conclusions:

• The CDDM framework effectively reconstructs high-quality sparse-view CT images.
• CDDM demonstrates significant computational efficiency and improved image clarity.
• This approach offers a promising solution for dose reduction in CT imaging.