Key result
AI-CVD phenotyping outperforms traditional Agatston scoring for detecting obstructive stenosis with ~0.94 AUC.
Why the study?
CAC scans quantify calcified plaques but do not directly measure non-calcified or high-risk plaques seen on CCTA. Whether AI-derived cardiometabolic biomarkers from CAC scans can predict these CCTA-defined plaque phenotypes and outperform CAC score alone was unknown.
Does comprehensive AI-based cardiometabolic phenotyping of CAC scans improve the discrimination of obstructive and high-risk plaques on CCTA compared to the traditional Agatston CAC score alone in asymptomatic adults?
Cross-Sectional (n=1,259)
Does comprehensive AI-based cardiometabolic phenotyping of CAC scans improve the discrimination of obstructive and high-risk plaques on CCTA compared to the traditional Agatston CAC score alone in asymptomatic adults?
AI-based cardiometabolic phenotyping of routine non-contrast CAC scans significantly outperforms traditional Agatston scoring in predicting obstructive, high-risk, and non-calcified plaques on CCTA, potentially enabling more precise risk-guided referral for contrast-enhanced imaging.
Background CAC scans quantify calcified plaques but do not directly measure non-calcified plaques (NCP) or high-risk plaques (HRP) described in CCTA reports. Whether AI-derived cardiometabolic biomarkers from CAC scans can predict CCTA-defined plaque phenotypes is unknown. Purpose To evaluate whether comprehensive AI-based phenotyping of CAC scans can discriminate obstructive and high-risk plaques, and whether these models outperform CAC score alone. Methods We applied FDA-cleared AI-CVD deep learning algorithms to CAC scans of 1,259 asymptomatic participants from the Miami Heart Study (aged 40 to 65 years, 48.7% female) who underwent paired CAC and CCTA imaging. AI-CVD algorithms extracted cardiometabolic biomarkers from CAC scans, including calcium scores, plaque burden indices, pericoronary adipose tissue textural features, and metabolic biomarkers such as epicardial and visceral adipose tissue volume, skeletal muscle mass and density, and liver attenuation within the available field of view of the CAC scans. Results AI-CVD demonstrated significant improvements over the traditional Agatston CAC score for detecting obstructive stenosis (AUC 0.943 vs. 0.881), HRP (AUC 0.779 vs. 0.626), and NCP (AUC 0.771 vs. 0.679). These findings suggest that non-contrast CAC imaging contains potential vascular and metabolic phenotype information beyond traditional Agatston scoring, enabling prediction of plaque characteristics not directly visualized on CAC scans. Conclusion Comprehensive AI-based cardiometabolic phenotyping of routine CAC scans enables cross-sectional prediction of obstructive and high-risk plaque phenotypes typically assessed only with contrast-enhanced CCTA. These findings suggest that non-contrast CAC imaging functions as a multidimensional cardiometabolic phenotyping tool, expanding its clinical utility beyond calcium quantification and potentially enabling more precise, risk-guided referral to CCTA.ROC curveROC curve
No takes yet. Share an insight, caveat, or question.
Atlas et al. (2026) conducted a cross-sectional in Subclinical coronary atherosclerosis (obstructive, high-risk, and non-calcified plaques) (n=1,259). AI-CVD deep learning cardiometabolic phenotyping of CAC scans vs. Traditional Agatston CAC score was evaluated on Discrimination of obstructive stenosis, high-risk plaques (HRP), and non-calcified plaques (NCP) on CCTA (AUC 0.943 vs 0.881 (obstructive stenosis); AUC 0.779 vs 0.626 (HRP); AUC 0.771 vs 0.679 (NCP)). AI-CVD phenotyping of CAC scans outperformed Agatston CAC score for detecting obstructive stenosis (AUC 0.943 vs 0.881), high-risk plaque (0.779 vs 0.626), and non-calcified plaque (0.771 vs 0.679).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: