Mammography-based deep learning model for coronary artery calcification.

Area of Science:

• Radiology
• Cardiology
• Artificial Intelligence

Background:

• Mammography is primarily used for breast cancer screening.
• Mammography has the potential to predict cardiovascular disease (CVD) risk.
• Coronary artery calcium (CAC) scoring is an established predictor of coronary events.

Purpose of the Study:

• To develop and evaluate deep learning models utilizing mammography images for predicting CAC scores.
• To assess the performance of these models in identifying individuals at risk for cardiovascular events.

Main Methods:

• A convolutional neural network (CNN) was developed using a training dataset of 5235 Korean women who underwent mammography and CAC computed tomography.
• The Resnet18 model, enhanced with contrastive learning strategies, was employed.
• Model performance was evaluated using sensitivity, specificity, AUROC, and accuracy.
• Age and menopausal status were incorporated to assess their impact on model performance.

Main Results:

• The CNN-based deep learning model achieved an AUROC of 0.761.
• Incorporating age and menopausal status improved the model's AUROC to 0.776.
• The deep learning model's performance was comparable to the Framingham risk score (AUROC 0.736).

Conclusions:

• Mammography-based deep learning models demonstrate significant potential for predicting CAC scores.
• These models offer a promising approach for dual-risk assessment of breast cancer and cardiovascular disease.
• Further validation in diverse populations and integration with national screening programs are recommended.