Fully Automated Artery-Specific Calcium Scoring Based on Machine Learning in Low-Dose Computed Tomography Screening

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Coronary artery calcium (CAC) scoring is crucial for cardiovascular risk assessment.
• Semi-automated software is currently the clinical standard for CAC quantification.
• Artery-specific scoring requires detailed analysis, which can be time-consuming.

Purpose of the Study:

• To evaluate a novel machine learning-based, fully automated software for artery-specific coronary artery calcium scoring.
• To compare the performance of the automated software against a semi-automated reference standard.

Main Methods:

• 505 patients underwent non-contrast-enhanced calcium scoring computed tomography (CSCT).
• A machine learning algorithm quantified Agatston score (AS), volume score (VS), and mass score (MS) for individual coronary arteries.
• Results were compared to a reference standard derived from semi-automated software analysis, assessing correlation, agreement, and evaluation time.

Main Results:

• The automated software demonstrated strong correlation (Spearman's rho > 0.969) and excellent agreement (ICC > 0.919) with the reference standard for AS, VS, and MS.
• Mean assessment time was significantly reduced from 59 seconds (semi-automated) to 5.9 seconds (automated).
• Risk class assignment accuracy was high, with 98.4% correct assignments.

Conclusions:

• The fully automated, artery-specific CAC scoring algorithm is a time-efficient procedure.
• It exhibits excellent correlation and agreement compared to the established semi-automated approach.
• This automated tool is suitable for clinical application in artery-specific calcium scoring.