Comprehensive Summary
Narang et al. tested a deep-learning system that guided nurses with no prior echocardiography experience to obtain limited transthoracic studies. In 240 adult studies at two U.S. centers, blinded echocardiographers determined that diagnostic adequacy was achieved in 98.8% of patients for LV size and function, 92.5% for RV size, and 98.8% for pericardial effusion, exceeding FDA thresholds. Compared with sonographers, adequacy was similar for most primary clinical parameters (LV size: 98.7% vs. 100%; RV size: 92.3% vs. 96.2%) but lower for some secondary parameters, especially IVC size (57.4% vs. 91.5%). Median acquisition time was 30 minutes, and diagnostic adequacy was consistent across patient BMI and pathology subgroups.
Outcomes and Implications
AI guidance enabled novices to capture diagnostic echocardiograms for basic parameters, suggesting a possibility of expanding imaging access in settings with few sonographers. This could improve triage and early detection, but generalizability is limited by the small, two-center trial, structured nurse training, and reduced adequacy for some views. The system only supports 10 standard views, and clinical outcomes or workflow-specific benefits are untested. Larger, multicenter studies are required before considering adoption of this deep-learning approach.