AI-driven calcium-omics for advanced cardiovascular risk stratification
Rui Castro, A S Barros, J M Pedrosa, F S Nunes, C Costa, R F CarvalhoAbstract
Background
Cardiovascular (CV) risk stratification in asymptomatic patients relies largely on classical CV risk factor identification and treatment. In the last decade, the detection and quantification of coronary artery calcium (CAC scoring), has emerged as a useful tool to further refine CV risk stratification, beyond traditional risk factors.
Purpose
Currently, CAC scoring impacts clinical decisions largely based on its absolute value – the higher the value, the higher the CV risk. However, we theorize that more information can be extracted from the CAC scoring scans, i.e, a calcium-omics approach. Thus, calcium-omics is related to all the features that may be extracted from CAC segmentations. We aimed to show that the addition of calcium-omics improves on CV risk prediction beyond tradicional risk factors and isolated CAC scoring.
Methods
A total of 1607 patients who underwent CAC scoring at our institution were restrospectively enrolled. From CAC segmentations, we extracted conventional CAC score and AI-derived calcium-omics features capturing quantitative characteristics of calcification burden, distribution, morphology, and texture. Demographic variables were recorded for all patients. Participants were followed for cardiovascular outcomes, including major adverse cardiovascular events (MACE) and all-cause mortality. Five prognostic models were developed: (1) demographic variables alone, (2) isolated CAC score, (3) CAC score + demographics, (4) calcium-omics alone, and (5) calcium-omics + demographics. Model performance was evaluated using the C-index, integrated Brier score (IBS), and area under the receiver operating characteristic curve (AUC) in training (70%) and testing (30%) datasets. Kaplan–Meier curves compared risk stratification based on CAC categories versus Cox model–derived risk groups.
Results
Model performance is summarized in Table 1. The combined calcium-omics + demographic model demonstrated the best predictive performance in the testing dataset (C-index 0.61, IBS 0.13, AUC 0.58). Kaplan–Meier analyses showed clearer separation of survival curves for Cox-derived risk groups compared with isolated CAC score categories, indicating improved risk stratification.
Conclusion
AI-driven calcium-omics enhances cardiovascular risk prediction beyond demographic variables and traditional CAC scoring, supporting the added value of advanced quantitative analysis of CAC scans for refined risk stratification.Table 1.Kaplan–Meier Analyses