Segmentation-Guided Accelerating Diffusion Model for Cardiac CT Motion Artifact Reduction via Limited-Angle Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Diagnostics

Background:

• Coronary computed tomography angiography (CCTA) is crucial for diagnosing cardiac disease but suffers from motion artifacts at high heart rates.
• Limited-angle CT (LA-CT) reduces acquisition time and motion artifacts but introduces severe wedge artifacts.
• Existing diffusion models for medical imaging have high computational costs, limiting clinical use.

Purpose of the Study:

• To develop a novel method for reconstructing motion-free cardiac CT images from LA-CT data.
• To suppress severe wedge artifacts inherent in LA-CT reconstruction.
• To improve the clinical applicability of diffusion models in cardiac imaging.

Main Methods:

• Proposed a Segmentation-Guided Accelerating Diffusion Model (SGADM) for LA-CT imaging.
• SGADM employs an innovative diffusion model for direct, high-quality CT image generation with reduced sampling steps (<10).
• Integrated diffusion perceptual loss for data distribution consistency and segmentation guidance for enhanced coronary artery accuracy.

Main Results:

• SGADM effectively reconstructs high-quality CCTA images with minimal motion artifacts.
• The model successfully suppresses wedge artifacts from LA-CT data.
• Quantitative and qualitative evaluations on simulated and real datasets confirmed SGADM's efficacy.

Conclusions:

• SGADM offers a computationally efficient and effective solution for motion- and artifact-free cardiac CT image reconstruction.
• The method significantly enhances the diagnostic quality of CCTA, especially in challenging cases (arrhythmias, high heart rates).
• SGADM shows strong potential for clinical translation in cardiovascular imaging.