AI could predict who will have a heart attack (www.technologyreview.com)

Startups including Bunkerhill Health, Nanox.AI and HeartLung Technologies are applying AI to routine chest CTs to automatically detect and quantify coronary artery calcium (CAC), a well-known marker of coronary plaque and heart-attack risk that often goes unreported when scans are done for trauma or lung screening. By mining the roughly 20 million chest CTs performed annually in the U.S., these algorithms could flag high CAC scores and surface patients who wouldn’t otherwise get targeted cardiac screening, potentially expanding access to risk stratification without a dedicated heart CT. Technically, these tools repurpose non-heart-gated CT images to estimate CAC scores that usually require specialized cardiac scans; CAC presence implies earlier, lipid-rich plaque may be present even if calcified plaque itself is more stable. But the approach is unproven at scale and raises major implementation, clinical and ethical questions: CAC screening hasn’t clearly reduced mortality in population studies, insurers largely don’t reimburse AI-derived scores, workflows to act on incidental findings are lacking, and automated reporting could drive overdiagnosis, unnecessary expensive therapies (e.g., PCSK9 inhibitors) or false reassurance. Critics warn this could shift how disease is defined—creating “machine-based nosology” and potential inequities if algorithm access and quality vary. Clinicians’ judgment and downstream care pathways remain crucial to realize benefit.