AI-Powered Coronary Calcium Measurement Streamlines Treatment Planning

Hagopian and colleagues developed a deep learning model that automatically calculates coronary artery calcium in routine chest computed tomography scans performed for non-cardiac reasons. In addition to predicting cardiovascular events, this AI system successfully identifies patients who may benefit from lipid-lowering therapy.

AI Model and Development

Hagopian and his team developed a specialized deep learning model to quantitatively assess coronary artery calcium in computed tomography scans at U.S. Veterans Affairs hospitals. This system automatically performs calcium segmentation on non-contrast and non-targeted chest computed tomography scans.

Accuracy Rates and Clinical Tests

The model was tested by comparing it with clinical electrocardiogram-tested calcium scores. In 795 matched scans, an accuracy rate of 89.4% was achieved in distinguishing between zero and non-zero calcium, while an 87.3% success rate was attained for scores below and above the threshold of 100.

Cardiovascular Risk Prediction

AI-derived calcium scores successfully predicted 10-year all-cause mortality and composite risks such as first myocardial infarction, stroke, or death. It was determined that risk rates significantly increased in high-score groups.

Lipid-Lowering Therapy Screening

In opportunistic screenings conducted on a separate cohort undergoing low-dose lung cancer screening, the AI model successfully identified patients who could benefit from lipid-lowering therapy. Expert cardiologists also confirmed that this patient group would benefit from the treatment.