Artificial Intelligence Measurements of Heart Scans Show Stronger Link to Patient Survival Than Clinician Assessments

Recent research published in The Lancet Digital Health reveals that artificial intelligence measurements of heart function derived from MRI scans are more strongly linked to patient survival rates than assessments performed by clinicians. The findings suggest that automated tools could eventually serve as a reliable first-line approach for analyzing critical cardiac data.

Understanding Ejection Fraction in Heart Care

The investigation centered on ejection fraction, widely considered the most vital measure of cardiac performance. Ejection fraction calculates how effectively the heart pumps blood and acts as a central clinical indicator for numerous treatments. Cardiologists rely on this metric to determine whether patients qualify for specialized heart failure therapies, implanted devices, and procedures designed to restore blood flow. Additionally, the measurement helps clinicians gauge surgical risks and monitor heart function during chemotherapy regimens.

Study Design and Patient Outcomes

In what is described as the largest study of its kind, researchers examined cardiac MRI scans from 27,382 patients. These individuals were undergoing evaluations for known or suspected heart conditions across medical centers in the United Kingdom and the United States. Participants were monitored for an average of more than five years, during which 3,458 deaths were recorded.

The study compared ejection fraction measurements produced by an automated AI tool against those generated by experienced clinicians evaluating the same cardiac MRI scans. According to the results, the AI-derived measurements were slightly, yet significantly, more accurate in predicting which patients faced an increased risk of mortality. These statistical outcomes remained consistent even after researchers adjusted for variables such as patient age and sex.

Speed and Efficiency of Automated Analysis

Beyond predictive capability, the AI software demonstrated significant efficiency. The technology processed and analyzed scans in approximately two to three seconds per patient. For comparison, previous evaluations by the research team indicated that expert human analysis typically requires around 13 minutes per scan.

Dr. George Joy, an NIHR clinical lecturer from the Cardiovascular Clinical Academic Group at City St George’s, University of London and lead author of the study, noted that ejection fraction is among the most essential metrics in cardiology. He stated that the findings demonstrate the capacity of AI to measure this marker with remarkable consistency and precision, leading to a small but meaningful enhancement in identifying high-risk individuals.

The Importance of Human Oversight

Despite the promising performance metrics, the study authors emphasize that clinical oversight remains indispensable. While automated errors were infrequent—affecting roughly 1% of patients at the individual level and translating to about 0.08% of total images analyzed—the researchers maintain that qualified professionals are necessary to verify outcomes and interpret findings within a broader diagnostic framework.

Dr. Rhodri Davies, an associate professor at University College London, senior author, and inventor of the AI tool, emphasized that even modest gains in accuracy can carry substantial weight when scaled across thousands of evaluations. He clarified that the technology is designed to support, rather than replace, medical professionals by delivering faster, uniform measurements at a lower cost to help streamline clinical decisions.

Future Directions

As healthcare systems encounter growing clinical demands, automated tools could potentially alleviate workloads while preserving high standards of patient care. The research team aims to expand on these findings by combining AI measurements from cardiac scans with data from other diagnostic tests, such as electrocardiograms, to identify muscle abnormalities earlier and initiate interventions before permanent damage occurs.

• Evaluated cardiac MRI scans from 27,382 patients across the U.K. and U.S.
• Tracked patient outcomes over an average of more than five years.
• Demonstrated AI analysis speeds of two to three seconds per patient.