AI Tool Matches Cardiologists in Echocardiography Interpretation Across 39 Tasks: Study Shows

Echocardiography remains a fundamental tool in the evaluation and monitoring of heart conditions, but the process is traditionally dependent on expert interpretation of numerous ultrasound video clips. Recognizing this limitation, the research team developed PanEcho using a multitask deep learning model to perform a wide range of diagnostic and measurement tasks.

The study involved training the AI system on a vast dataset comprising 1.2 million echocardiographic videos from 32,265 TTE studies involving over 24,000 patients at Yale New Haven Health System (YNHHS) facilities, collected between 2016 and 2022. The model was subjected to both internal and external validations, including tests on real-world point-of-care ultrasound data.

The study led to the following findings:

• In internal validation, the AI system PanEcho completed 18 diagnostic classification tasks with a median AUC of 0.91.
• For measurement tasks such as estimating ejection fraction and chamber size, it achieved a median normalized mean absolute error of 0.13.
• The system estimated left ventricular ejection fraction with a mean absolute error of 4.2% in internal testing and 4.5% in external validation.
• It accurately identified moderate or worse left ventricular systolic dysfunction, achieving an AUC of 0.99.
• PanEcho detected severe aortic stenosis with perfect accuracy in external validation, achieving an AUC of 1.00.
• The model also performed well in identifying right ventricular systolic dysfunction, with an external AUC of 0.94.
• In abbreviated transthoracic echocardiography protocols, the AI maintained strong performance, achieving a median AUC of 0.91 across 15 tasks.
• On point-of-care ultrasound scans from emergency departments, it sustained robust results with a median AUC of 0.85 across 14 diagnostic tasks.

These findings suggest that PanEcho could significantly enhance echocardiographic workflows by serving as an adjunct reader in clinical labs or as a reliable tool for cardiovascular screening in environments with limited access to trained cardiologists. The authors emphasize, however, that prospective evaluation in clinical settings is necessary before the system can be broadly integrated into routine care.

By automating a traditionally manual and time-intensive process, this AI-powered system offers a promising avenue to expand access to cardiac diagnostics and reduce bottlenecks in care delivery, particularly in underserved regions.

Reference:

Holste G, Oikonomou EK, Tokodi M, Kovács A, Wang Z, Khera R. Complete AI-Enabled Echocardiography Interpretation With Multitask Deep Learning. JAMA. Published online June 23, 2025. doi:10.1001/jama.2025.8731