Comprehensive Summary
He et al. conducted a blinded randomized non-inferiority trial comparing AI-based versus sonographer-based initial assessment of left ventricular ejection fraction (LVEF) from echocardiograms in 3,495 echocardiographic studies. Studies were randomized 1:1, with cardiologists providing final adjudication blinded to the initial method. Substantial changes (>5% LVEF difference) between initial and final assessments occurred less frequently with AI (16.8%) than with sonographers (27.2%), demonstrating both non-inferiority and superiority. The mean absolute difference between final and initial assessments was 2.79% and 3.77% for the AI-based method and sonographers respectively. AI also shortened annotation time for both sonographers and cardiologists and cardiologists were unable to reliably identify the source of the initial assessment. Across subgroups, AI yielded more consistent results and aligned more closely with historical cardiologist assessments, supporting its reproducibility and precision.
Outcomes and Implications
AI LVEF assessment offers a quicker and more reproducible alternative to traditional sonographers, potentially reducing human variability and improving echocardiographic analysis. By reducing re-interpretation rates and shortening reading times, AI can enhance efficiency in healthcare settings with high volume. These findings suggest AI may help standardize cardiac function measurement and support therapeutic decision-making. More extensive validation across different patient populations, imaging protocols, and healthcare settings is needed before considering adoption. If proven effective in the aforementioned conditions, AI-based LVEF assessment could improve consistency, reduce workload, and support better cardiac care.