Deep convolutional neural networks to predict cardiovascular risk from computed tomography

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence

Background:

• Coronary artery calcium (CAC) on computed tomography (CT) predicts cardiovascular events.
• Current CAC quantification is time-consuming, requires expertise, and specialized equipment, limiting routine clinical use.

Purpose of the Study:

• To develop and validate a robust, time-efficient deep learning system for automated CAC quantification on routine CT scans.
• To assess the predictive value of automated CAC scoring for cardiovascular events.

Main Methods:

• A deep learning system was developed to automatically quantify CAC on cardiac-gated and non-gated chest CT scans.
• The system was evaluated in 20,084 individuals across asymptomatic and chest pain cohorts (Framingham Heart Study, NLST, PROMISE, ROMICAT-II).

Main Results:

• The automated CAC score strongly predicted cardiovascular events (multivariable-adjusted hazard ratios up to 4.3), independent of traditional risk factors.
• High correlation with manual quantification and robust test-retest reliability were demonstrated.
• The system proved effective across diverse patient cohorts.

Conclusions:

• The deep learning system provides a clinically valuable, automated method for CAC quantification.
• Automating this proven imaging biomarker can guide cardiovascular disease management and improve population health outcomes.