Enhancing cardiovascular risk prediction through AI-enabled calcium-omics

Area of Science:

• Cardiovascular Imaging
• Radiomics
• Predictive Analytics

Background:

• The Agatston score is a standard but limited predictor of major adverse cardiovascular events (MACE).
• It does not capture detailed calcification features crucial for understanding coronary artery disease pathophysiology.
• Individualized risk assessment requires analysis of calcification characteristics like number, distribution, and density.

Purpose of the Study:

• To develop and validate a novel risk prediction model using detailed coronary artery calcification features (calcium-omics).
• To compare the performance of the calcium-omics model against the traditional Agatston score for MACE prediction.
• To identify specific calcium-omics features that are significant determinants of MACE risk.

Main Methods:

• Utilized hand-crafted calcification features (calcium-omics) from 2457 CT calcium score (CTCS) images.
• Employed Cox time-to-event modeling, elastic net, and synthetic sampling for imbalanced data.
• Developed and validated a calcium-omics predictive model using an 80/20 training/testing split.

Main Results:

• Calcium-omics features, including number of calcifications, LAD mass, and diffusivity, were key risk determinants.
• Dense calcifications (>1000 HU) were associated with reduced MACE risk.
• The calcium-omics model achieved superior performance (C-index 80.5%/71.6%, 2-year AUC 82.4%/74.8%) compared to Agatston, identifying 13.2% more MACE cases.

Conclusions:

• Calcium-omics offers improved risk prediction for MACE compared to the conventional Agatston score.
• This novel approach enhances identification of patients requiring intensive cardiovascular follow-up and treatment.
• Further multi-institutional studies are warranted to validate and implement calcium-omics in clinical practice.