AI heart fat scans: A sharper risk predictor—with limits

Mayo Clinic’s latest research doesn’t just tweak cardiovascular risk prediction—it targets a blind spot in standard imaging. By training AI to quantify pericardial fat (the fat surrounding the heart) from routine CT scans, the team demonstrated a 15–20% improvement in accuracy over traditional models like the Framingham Risk Score. That matters because heart disease kills nearly 18 million people annually, and current tools miss high-risk patients who don’t fit classic profiles—like lean individuals with hidden fat deposits.

The study, published in The Lancet Digital Health, analyzed scans from 2,500+ patients across five years, making it one of the larger AI-imaging trials to date. But its strength—retrospective data from a single health system—is also its constraint. Observational studies can show associations, not causation, and Mayo’s cohort was predominantly white and middle-aged. That leaves open whether the AI’s accuracy holds for younger adults, diverse ethnic groups, or patients with existing heart conditions.

Critically, this isn’t about replacing doctors’ judgment. The AI doesn’t diagnose; it refines a probability score. As Dr. Erik B. Schelbert, the study’s senior author, noted, the goal is to ‘identify individuals who might benefit from earlier, more aggressive prevention’—not to trigger unnecessary interventions.

For patients today, the practical impact is zero. The AI tool isn’t FDA-approved, and Mayo hasn’t announced plans to commercialize it. Even if it were available, most primary care clinics lack the high-resolution CT scanners required. The study’s real significance lies in its validation of a long-suspected link: pericardial fat isn’t just a marker of risk—it’s an active contributor to inflammation and plaque buildup. That distinction could eventually reshape guidelines, but current ACC/AHA recommendations still prioritize cholesterol, blood pressure, and smoking status.

What’s missing? Prospective trials testing whether acting on these AI-generated risk scores actually improves outcomes. A 2021 meta-analysis in JAMA Cardiology found that even advanced imaging tools rarely change patient behavior without paired interventions like statin therapy or lifestyle programs. And then there’s cost: Adding AI analysis to every cardiac CT could inflate healthcare spending without clear proof it saves lives.

The Mayo team is now collaborating with NHLBI-funded researchers to validate the tool in broader populations. But the timeline for clinical adoption—if it happens—is measured in years, not months.