OPPORTUNISTIC ASSESSMENT OF CARDIOVASCULAR RISK USING AI-DERIVED STRUCTURAL AORTIC AND CARDIAC PHENOTYPES FROM...

Area of Science:

• Cardiovascular imaging and AI
• Radiomics for risk prediction
• Preventive cardiology

Background:

• Cardiovascular disease (CVD) primary prevention relies on risk scores (PCE, PREVENT).
• These scores often lack necessary data in electronic health records (EHR).
• Routinely collected data, like chest CTs, may enhance risk prediction.

Purpose of the Study:

• To evaluate a radiomics model using chest CT features for predicting major adverse cardiovascular events (MACE).
• To determine if this model adds value to existing clinical risk algorithms (PCE, PREVENT).
• To assess performance in patients with incomplete data for traditional scores.

Main Methods:

• A LASSO model was developed using cardiac and aorta radiomics features from 13,437 lung cancer screening CTs.
• The model predicted fatal MACE over 12 years.
• External validation was performed on 4,303 individuals, comparing the radiomics score to PCE and PREVENT scores.

Main Results:

• The radiomics score significantly improved MACE prediction compared to PCE (c-index 0.653 vs. 0.567).
• Performance was consistent even with missing input variables for traditional scores.
• Statin eligibility based on the radiomics score identified higher MACE incidence.

Conclusions:

• A cardiac shape-based radiomics model from chest CT predicts cardiovascular events beyond clinical algorithms.
• The model performs well even when traditional risk calculator inputs are missing.
• High-risk individuals identified by the radiomics score may benefit from intensified primary prevention strategies.