Deep learning to predict cardiovascular mortality from aortic disease in heavy smokers

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Radiomics

Background:

• Aortic angiopathy is a key indicator of cardiovascular disease (CVD) burden.
• Current risk prediction relies mainly on maximum aortic diameter, with other features' prognostic value uncertain.
• Heavy smokers represent a high-risk population for CVD mortality.

Purpose of the Study:

• To develop a deep learning framework for automated quantification of thoracic aortic disease features.
• To assess the prognostic value of these deep learning-derived features in predicting CVD mortality.
• To evaluate the utility of these features in a high-risk population of heavy smokers.

Main Methods:

• Utilized non-contrast chest CT scans from the National Lung Screening Trial (NLST).
• Developed a deep learning framework to quantify aortic maximum diameter, volume, and calcification burden.
• Analyzed associations between quantified aortic features and CVD mortality in 24,770 participants with a mean follow-up of 6.3 years.

Main Results:

• Aortic calcification burden and aortic volume were independently associated with CVD mortality.
• These associations remained significant after adjusting for traditional CVD risk factors and coronary artery calcifications.
• Deep learning-derived aortic features demonstrated prognostic value beyond maximum diameter.

Conclusions:

• Deep learning-based quantification of aortic features offers valuable prognostic information for CVD mortality.
• Aortic calcifications and volume are significant predictors of CVD mortality in heavy smokers.
• Incorporating these novel features may enhance CVD risk prediction and personalize prevention strategies in high-risk groups.