Predicting cardiovascular events from routine mammograms using machine learning

Area of Science:

• Artificial Intelligence in Medicine
• Cardiology
• Radiology

Background:

• Cardiovascular disease (CVD) risk is often underestimated in women.
• Midlife women undergoing screening mammography are at increased risk for CVD.
• Mammographic features like breast arterial calcification and density correlate with CVD risk.

Purpose of the Study:

• To develop and validate a deep learning algorithm for cardiovascular risk prediction using mammography images.
• To assess the performance of the algorithm against established cardiovascular risk assessment tools.

Main Methods:

• A deep learning model (DeepSurv) was developed using mammography images from the Lifepool cohort.
• The model predicted major cardiovascular events.
• Performance was evaluated using the concordance index and compared to standard risk models (e.g., PREDICT, PREVENT).

Main Results:

• The study included 49,196 women with a median follow-up of 8.8 years.
• The DeepSurv model achieved a concordance index of 0.72 (95% CI 0.71-0.73).
• Performance was comparable to traditional risk prediction models incorporating age and clinical data.

Conclusions:

• A deep learning algorithm utilizing mammographic features and age can predict cardiovascular risk effectively.
• Mammography-based risk assessment presents a potential new avenue for cardiovascular screening in women.
• This AI approach may improve early detection and management of cardiovascular risk in this population.