Artificial Intelligence in Cardiac Point-of-Care Ultrasound: A Narrative Review.
Authors
Affiliations (5)
Affiliations (5)
- Department of Emergency Medicine, Hadassah University Hospital-Ein Kerem, Jerusalem 91120, Israel.
- Faculty of Medicine, Hebrew University of Jerusalem, Jerusalem 91905, Israel.
- Department of Emergency Medicine, Northwell, New Hyde Park, New York, NY 11040, USA.
- Department of Emergency Medicine, Staten Island University Hospital, Staten Island, NY 10305, USA.
- Cardiology Department, Jesselson Integrated Heart Center, Eisenberg R&D Authority, Shaare Zedek Medical Center, Jerusalem 91031, Israel.
Abstract
<b>Background</b>: Cardiac point-of-care ultrasound (POCUS) is widely used in emergency and acute care settings. Still, broader use remains limited by operator dependence and variability in image acquisition and interpretation. Artificial intelligence (AI), including machine learning and deep learning methods, has been applied to cardiac POCUS to support image acquisition, automate quantitative measurements, and assist interpretation. <b>Methods</b>: We performed a narrative review of current applications of AI-assisted cardiac POCUS. A targeted literature search of PubMed and Google Scholar from 2018 to 2026 was conducted using terms related to AI, machine learning, deep learning, and cardiac ultrasound. Studies evaluating AI-assisted cardiac ultrasound in clinical, educational, or image-acquisition settings were included, with emphasis on recent, clinically relevant applications. <b>Results</b>: The most developed application of AI-assisted cardiac POCUS is an automated assessment of left ventricular systolic function, particularly the left ventricular ejection fraction (LVEF), where multiple studies report agreement with expert interpretation or formal echocardiography and improved performance among novice users. AI-assisted tools have also been evaluated for pericardial effusion detection, guidance for image acquisition, and education. More complex applications, including diastolic function assessment and hemodynamic measurements such as LVOT-VTI, remain less well validated and more dependent on image quality. Across studies, performance is closely linked to image acquisition quality and has often been evaluated under controlled rather than real-world conditions. <b>Conclusions</b>: Current evidence supports AI-assisted cardiac POCUS primarily as a decision-support tool, with the strongest data for automated assessment of LVEF. Other applications remain investigational.