An artificial intelligence system that measures heart function from a cardiac MRI in just two to three seconds has demonstrated a slight but statistically significant edge over expert clinicians in predicting patient mortality. The findings, published on November 2, 2026, in The Lancet Digital Health, stem from a collaboration led by City St George’s, University of London and University College London (UCL), drawing on data from 27,382 patients across the United Kingdom and the United States.
- Speed advantage: The AI completes each MRI analysis in roughly 2-3 seconds, compared with about 13 minutes for a human expert.
- Predictive edge: AI-derived ejection fraction predicted death risk slightly but significantly better than clinician measurements, even after adjusting for age and sex.
- Scale of evidence: The retrospective study tracked 27,382 patients over a median follow-up exceeding five years, during which 3,458 deaths occurred.
What the AI Measures and Why It Matters
The tool focuses on ejection fraction — the proportion of blood the heart pumps with each beat — which is a cornerstone indicator of cardiac pumping capacity. Clinicians rely on this metric to guide decisions ranging from surgical risk assessment to monitoring patients undergoing chemotherapy. According to the study, the AI-calculated ejection fraction predicted mortality risk “slightly but significantly” better than human-derived measurements.
The speed differential is stark. Prior work from the same team indicates that expert analysis of a single cardiac MRI typically takes around 13 minutes, while the AI processes each scan in seconds. Lead author Dr. George Joy of City St George’s said the findings suggest the AI “can measure this important marker with remarkable consistency and accuracy.”
“AI is not replacing clinicians, but it can offer faster and more consistent measurements at low cost.” — Dr. Rhodri Davies, UCL, senior author and inventor of the tool
Davies acknowledged that the improvement is modest, but stressed that “even modest improvements in accuracy can have a significant impact when applied to thousands of patients.” Error rates were low: approximately 1% at the patient level, affecting 0.08% of images.
Human Oversight Remains Essential
The authors are explicit that the technology is designed to augment, not replace, clinical expertise. Clinicians must still verify results and interpret MRIs within their full clinical context. The study itself is retrospective — analyzing existing data rather than running a real-time trial — and the researchers emphasize that validation in practice is a necessary next step.
The team’s forward-looking plan involves combining these AI measurements with other diagnostic inputs, particularly electrocardiograms, to detect cardiac muscle problems earlier. This aligns with a broader wave of AI integration in medical imaging, where tools are increasingly tasked with routine measurements that demand consistency and speed.
The Consistency Argument
The most compelling aspect of this research is not the headline-grabbing notion of AI outperforming doctors, but the quieter, more clinically relevant benefit: consistency. Two cardiologists measuring the same MRI can produce somewhat different figures, and that variability carries clinical consequences. An AI that eliminates this noise offers a tangible advantage — not by making better judgments than a physician, but by making the same measurement the same way every time.
The honest framing matters here. The survival-prediction improvement is slight. The study is retrospective. Real-world validation is still pending. But the direction is sound, and the positioning is healthy: AI as a precision instrument serving the physician, not supplanting them. If the future of medicine looks like this — tools that free up 13 minutes per MRI so specialists can spend that time with patients — the potential is significant. The challenge, as always, will be ensuring that time savings translate into more human attention rather than simply more patients per hour.