AI Outperforms Clinicians in Assessing Heart Function From MRI

New research indicates that artificial intelligence can more accurately assess heart function from magnetic resonance imaging than expert clinicians, with stronger implications for predicting long-term survival. A study published online Nov. 2 in The Lancet Digital Health, led by George Joy, Ph.D., of City St. George's University of London, examined whether deep-learning AI segmentation of ejection fraction from cardiac MRI could outperform clinician-based measurements in predicting all-cause mortality. The researchers analyzed 27,382 consecutive patients with baseline cardiac MRI scans. Over an average follow-up period of 5.6 years, AI-derived ejection fraction demonstrated a better model fit than clinician-derived measurements, with a difference in Akaike information criterion of 161 favoring AI. It also showed a marginal improvement in discrimination, with a change in C index of 0.033. The study found that adding AI ejection fraction to clinician ejection fraction significantly improved predictive model fit, with a chi-square value of 168. In contrast, adding clinician ejection fraction to the AI model produced only a modest improvement, with a chi-square value of 7.5. After adjusting for age and sex, AI-derived ejection fraction continued to show better model fit, with a ΔAIC of 59, and marginally improved discrimination, with a ΔC index of 0.004. The findings suggest that AI can measure ejection fraction with greater consistency and precision, offering a small but meaningful enhancement in identifying patients at higher risk. The results may support broader integration of AI in cardiac MRI interpretation, particularly for risk stratification and long-term patient management.