Machine learning for coronary artery calcification detection and labeling using only native computer tomography

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cardiovascular disease (CVD) mortality is rising globally.
• Coronary artery calcification (CAC) is a key indicator of CVD events.
• Individual artery calcium scoring offers greater prognostic value than overall scores.

Purpose of the Study:

• To develop and evaluate a machine learning approach for segmenting, labeling, and quantifying coronary artery calcification.
• To utilize native coronary computed tomography (CT) for calcium analysis.
• To assess the accuracy of machine learning in CAC detection and localization.

Main Methods:

• A semi-automatic system with preprocessing to define the region of interest.
• Two random forest classifiers for CAC detection (vs. noise) and labeling (by artery).
• Calculation of Agatston and volume scores for individual calcifications, arteries, and the entire coronary tree.

Main Results:

• High accuracy achieved: 99.98% for CAC detection and 100% for CAC labeling.
• The developed method demonstrated promising results comparable to existing literature.
• Precise quantification of calcium scores at various levels (individual, arterial, total).

Conclusions:

• Machine learning algorithms can effectively segment, label, and quantify coronary artery calcification from native coronary CT.
• The high accuracy of this method supports its potential for improved cardiovascular risk stratification.
• This automated approach offers a valuable tool for clinical assessment of coronary artery disease severity.