Scoring of Coronary Artery Disease Characteristics on Coronary CT Angiograms by Using Machine Learning

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiomics

Background:

• Coronary CT angiography (CCTA) offers prognostic insights, but optimal data extraction methods are needed.
• Identifying patients at risk for adverse cardiovascular events is crucial for timely intervention.

Purpose of the Study:

• To develop and validate a machine learning (ML) model using CCTA vessel features to predict patient outcomes.
• To compare the performance of the ML model against conventional risk scores, including CAD-RADS.

Main Methods:

• Radiologists analyzed CCTA data, extracting four features for each of 16 coronary segments.
• Four ML model types were explored; performance was evaluated using the area under the receiver operating characteristic curve (AUC).
• Outcomes included all-cause mortality, coronary heart disease deaths, and nonfatal myocardial infarctions, assessed via the National Death Index.

Main Results:

• The ML model (k-nearest neighbors) demonstrated superior discrimination for all-cause mortality (AUC 0.77 vs. 0.72 for CAD-RADS, P < .001).
• For coronary artery disease deaths, the ML model achieved a higher AUC (0.85 vs. 0.79 for CAD-RADS, P < .001).
• The ML score improved treatment targeting for statin therapy, identifying 93% of patients with events compared to 69% with CAD-RADS.

Conclusions:

• Machine learning methods applied to CCTA vessel features significantly outperform conventional scores in discriminating patients with subsequent adverse cardiovascular events.
• The developed ML model offers a more accurate prognostic tool for cardiovascular risk stratification.
• These findings support the integration of ML-based radiomics into clinical practice for enhanced patient management.