Deep learning for vessel-specific coronary artery calcium scoring: validation on a multi-centre dataset.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Coronary artery calcium (CAC) scoring is crucial for cardiovascular risk assessment.
• Accurate, automated CAC scoring can improve efficiency and consistency in clinical practice.

Purpose of the Study:

• To develop and validate a fully automated deep learning (DL) algorithm for branch-wise CAC scoring.
• To assess the algorithm's performance on a multi-center dataset.

Main Methods:

• Retrospective analysis of 1171 patients undergoing CAC computed tomography.
• Automated CAC scoring using a DL software, providing total and branch-specific scores (RCA, LM, LAD, CX).
• Comparison with manual scoring by three readers on 300 cases, using consensus as the reference standard.

Main Results:

• The DL algorithm achieved 93% accuracy for overall risk class assignment.
• Detection of non-zero coronary calcium showed 97% sensitivity, 93% specificity, and 95% accuracy.
• Branch-specific classification accuracy was high (94% overall), with excellent performance for RCA (100%).

Conclusions:

• Fully automated total and vessel-specific CAC scoring using DL is feasible and accurate.
• The algorithm shows high agreement with manual assessments and reproducible results for vessel-specific scoring.
• This automated approach has the potential to enhance cardiovascular risk stratification.