Automatic Coronary Artery Plaque Quantification and CAD-RADS Prediction Using Mesh Priors

Area of Science:

• Medical Imaging
• Cardiovascular Research
• Computational Anatomy

Background:

• Coronary artery disease (CAD) is a major global health concern.
• Coronary CT angiography (CCTA) is crucial for diagnosing CAD and assessing cardiovascular event risk.
• Current analysis methods for CCTA, including CAD-RADS scoring, require accurate segmentation of coronary arteries and atherosclerotic plaque.

Purpose of the Study:

• To develop a novel method for directly inferring surface meshes of coronary artery lumen and plaque from CCTA.
• To utilize these surface meshes for automated CAD-RADS categorization.
• To evaluate the accuracy and feasibility of the proposed mesh-inference method for clinical application.

Main Methods:

• A novel approach was developed to directly infer surface meshes for coronary artery lumen and plaque using a centerline prior.
• The method was trained and evaluated on 2407 CCTA scans.
• Downstream task of CAD-RADS scoring was performed using the inferred meshes.

Main Results:

• The method achieved high lesion-wise volume intraclass correlation coefficients: 0.98 (calcified), 0.79 (non-calcified), and 0.85 (total plaque).
• Patient-level CAD-RADS categorization achieved a linearly weighted kappa (κ) of 0.75 on a 300-scan test set.
• Cross-institutional validation on 658 scans yielded a κ of 0.71, demonstrating robustness.

Conclusions:

• Direct inference of coronary artery lumen and plaque surface meshes from CCTA is feasible.
• This mesh-based approach enables automated and accurate CAD-RADS categorization.
• The method holds potential for improving clinical decision-making in patients with suspected CAD.