Automated coronary calcium detection and scoring on multicenter, multiprotocol noncontrast CT

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) is a primary global cause of death.
• Accurate assessment of coronary artery calcified plaques is crucial for cardiovascular risk stratification.
• Current methods for plaque detection and scoring can be labor-intensive and subjective.

Purpose of the Study:

• To develop and validate a deep learning method for automatic detection and scoring of coronary artery calcified plaques on noncontrast CT scans.
• To improve the accuracy and efficiency of cardiovascular risk assessment through automated plaque analysis.

Main Methods:

• Utilized five datasets (1 internal, 4 external) with 641 training and 160 testing noncontrast CT scans.
• Developed a deep learning model using the nnU-Net framework with simultaneous segmentation of aorta, heart, and lungs.
• Automatically computed Agatston scores for plaque burden quantification and compared with a previous method.

Main Results:

• The model achieved a strong correlation (r²=0.973) between predicted and reference Agatston scores.
• Demonstrated high performance with 89.3% precision, 89.1% recall, and 75.0% Dice score.
• Achieved 92.0% accuracy and Cohen's Kappa of 0.913 for Agatston group stratification, correlating with clinical outcomes.

Conclusions:

• The proposed nnU-Net-based deep learning method accurately detects and segments coronary artery plaques on noncontrast CT scans.
• Automated Agatston score calculation enables effective cardiovascular risk stratification.
• This approach facilitates opportunistic screening and large-scale population-based studies for CAD.