Synthesis of coronary 4D CT Image by denoising diffusion probabilistic model

Area of Science:

• Medical Imaging
• Computational Fluid Dynamics
• Artificial Intelligence

Background:

• Pressure drop fluctuations during the cardiac cycle offer prognostic insights for coronary artery disease (CAD).
• Traditional 4D computed tomography (CT) for time-variant flow analysis involves significant radiation exposure.
• A need exists for methods that provide dynamic hemodynamic information with reduced radiation doses.

Purpose of the Study:

• To develop a novel diffusion-based framework for synthesizing physiologically consistent 4D CT images.
• To perform 4D CT flow analysis using these synthesized images for improved CAD diagnosis.
• To reduce radiation exposure associated with traditional 4D CT flow analysis.

Main Methods:

• Utilized a denoising diffusion probabilistic model (DDPM) with a deformation module for anatomical reconstruction.
• Employed a computational fluid dynamics (CFD) model coupled with quasi-steady fluid-structure interaction (FSI) for 4D hemodynamic flow field calculation.
• Synthesized 4D CT images and performed flow analysis to assess diagnostic accuracy and radiation reduction.

Main Results:

• Achieved high image quality metrics: peak signal-to-noise ratio of 32.01 and structural similarity index of 0.937.
• Demonstrated excellent segmentation accuracy with an average Dice coefficient of 0.973.
• Calculated fractional flow reserve (FFR) with 90.5% accuracy, confirming diagnostic efficacy and reduced radiation exposure.

Conclusions:

• The synthesized 4D CT-based hemodynamic approach provides crucial time-variant information for CAD diagnosis.
• This method aids clinical decision-making by offering prognostic insights through dynamic lumen evaluation.
• The framework effectively reduces radiation exposure without compromising diagnostic quality, paving the way for safer CAD assessment.