Year: 2026
Type: article
Study design: Observational Study
Abstract: Background: Artificial intelligence (AI) may improve ultrasound image acquisition and interpretation, but there have been few validation studies comparing AI-enabled machines to traditional measurement workflows. In this diagnostic comparison study, we evaluated AI-enabled point of care ultrasound (POCUS) compared to standard transthoracic echocardiography (TTE) measurements. Methods: Hospitalized adults at a single academic center who underwent both AI POCUS and standard TTE scans within 7 days of one another were eligible for inclusion. The AI POCUS software used convolutional neural network models to estimate left ventricular ejection fraction (LVEF), right ventricular fractional area change (RVFAC), valvular regurgitation, valvular stenosis, and inferior vena cava (IVC) diameter based on standard cardiac views. Agreement between AI POCUS and TTE was assessed using weighted Cohen’s κ and exploratory Bland-Altman analyses. Results: Seventy patients were included, with a median of 1 day (IQR 1–2 days, range 0–6 days) between AI POCUS and TTE examinations. AI POCUS demonstrated substantial agreement with TTE for LVEF (κ = 0.75, 95% CI 0.42–0.92). Agreement was lower for RVFAC (κ = -0.003), IVC diameter (κ = 0.382), aortic regurgitation (κ = 0.541), mitral regurgitation (κ = 0.438), tricuspid regurgitation (κ = 0.400), and aortic stenosis (κ = 0.086). Using a dichotomized threshold of LVEF <50% versus ≥50%, AI POCUS demonstrated 80.0% sensitivity and 88.6% specificity for detecting reduced LVEF. Conclusions: AI POCUS demonstrated substantial agreement with TTE for the estimation of LVEF. Agreement was lower for other echocardiographic parameters. These findings support the potential of AI POCUS for assessing left ventricular function.
Source: POCUS Journal
Authors: Evan Baum(Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), John Kugler(Data curation, Project administration, Writing – original draft, Writing – review & editing), Michelle Lo(Data curation, Writing – original draft, Writing – review & editing), Caitlin Parmer-Chow(Data curation, Writing – original draft, Writing – review & editing), Vishnu Nair(Data curation, Writing – original draft, Writing – review & editing), Andre Kumar(Conceptualization, Data curation, Methodology, Supervision, Writing – original draft, Writing – review & editing)
Institutions: Stanford Medicine, Stanford University
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7236567452
