A Novel Fully Automated Deep Learning Model for Coronary Artery Calcification Detection on Computed Tomography

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) is a leading cause of mortality.
• Accurate detection and quantification of coronary artery calcifications (CACs) are crucial for risk stratification.
• Current methods for CAC analysis can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To evaluate a fully automated deep learning (DL) model for coronary artery segmentation and calcification detection.
• To assess the diagnostic accuracy of the DL model on non-contrast, non-gated CT scans.
• To determine the model's ability to predict coronary artery disease severity based on calcification volume.

Main Methods:

• A two-stage 3D segmentation pipeline was developed for coronary artery identification and calcification detection.
• The DL model was trained using anatomically refined labels and region-based optimization.
• Performance was evaluated against manual annotations in a large cohort (473 scans) and visually assessed by expert radiologists.

Main Results:

• The DL model demonstrated excellent performance in visual assessments and strong agreement with manual reference standards.
• High accuracy was achieved for coronary artery segmentation (κ 0.68–0.81) and calcification detection (κ 0.79–0.85).
• The model achieved high diagnostic accuracy for calcification detection (sensitivity 95%, specificity 98%) and correlated well with radiologist-reported disease severity.

Conclusions:

• The developed DL model accurately segments coronary arteries and detects calcifications on non-contrast CT.
• The model shows high diagnostic accuracy and effectively predicts coronary artery disease severity.
• This automated approach holds promise for efficient and reliable CAD assessment.