This study aims to apply deep learning models to chest x-rays (CXRs) to identify patients with a high coronary artery calcification (CAC) burden.
We proposed a novel multi-task learning framework and trained a model using 2,121 patients with paired gated CT scans and CXR images internally (Mayo Clinic) with CAC scores (0, 1-99, and 100+) as ground truths.
Results from the internal training were validated on multiple external datasets (EUH and VGHTPE) with significant racial and ethnic differences.
For the clinically relevant risk identification, the performance of our model on the internal and two external datasets reached AUCROCs of 0.86±0.02, 0.77±0.03, and 0.82±0.03 for 0 vs 400+ respectively.
For 0 vs 100+, we achieved AUCROCs of 0.83±0.03, 0.71±0.02, and 0.78±0.01 respectively.
In summary, the AI-CXR model, with its ability to identify patients with high coronary calcium burden from routine chest X-rays, offers a viable and cost-effective screening tool for opportunistic cardiovascular risk assessment.
The proposed CXR-CAC model demonstrates the potential to extract calcification features from a wide population variation (internal, external, foreign, and prospective) and classify CAC scores into clinically relevant categories.
This model represents a step towards harnessing the capabilities of AI and CXR imaging for cardiovascular risk assessment.
【MORE】