What is the clinical evidence from this study?

Study design: Observational. Population: Echocardiography / Left ventricular ejection fraction (LVEF) assessment (n=36676). Intervention: View-flexible deep learning framework vs. Clinically-calculated LVEF (standard of care). Primary outcome: LVEF estimation and classification of reduced LVEF (≤40%) (AUC 0.981).

What does this research mean for the field?

A view-flexible deep learning framework accurately estimates left ventricular ejection fraction from handheld cardiac ultrasound images, achieving an r2 of 0.80 and an AUC of 0.981, with comparable performance between expert and novice users. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The study aims to develop a deep learning framework to automatically analyze echocardiography images and estimate patient-related metrics.

April 3, 2026npj Cardiovascular Health0 citationsOpen Access

A view-flexible deep learning framework for automated analysis of 2D echocardiography

DAD. M. AnisuzzamanMayo Clinic in Arizona JMJeffrey G. MalinsCardiac Imaging JJJohn I. JacksonMayo Clinic

Key Result

A view-flexible deep learning framework accurately estimated left ventricular ejection fraction from handheld cardiac ultrasound images, achieving an r2 of 0.80 and an AUC of 0.981.

Key Points

The study aims to develop a deep learning framework to automatically analyze echocardiography images and estimate patient-related metrics.
Developed a view-flexible deep learning model for LVEF, age, and sex estimation.
Evaluated model performance across retrospective transthoracic echocardiography datasets.
Compared performance between prospective handheld cardiac ultrasound and transthoracic echocardiography data collected by experts and novices.
Model showed strong performance across datasets with LVEF r2 values of 0.80 for HCU and 0.86 for TTE.
AUC for sex classification was high at 0.985 in HCU versus 0.996 in TTE.
LVEF AUC for novice users was 0.968 compared to 0.983 for experts.

Study Design

Type

Observational (n=36,676)

Multicenter

Yes

Structured PICO

Does a view-flexible deep learning framework accurately estimate LVEF, age, and sex from 2D echocardiography across different devices and user expertise levels?

Population

33,924 patients undergoing 2D echocardiography, comprising a retrospective training cohort (n=19,627 from Mayo Clinic), retrospective testing cohorts (n=1,695 from Arizona; n=1,862 from Florida; n=10,015 from EchoNet Dynamic), a prospective TTE-HCU cohort (n=625), and a prospective HCU expert vs novice cohort (n=100).

Intervention

A view-flexible deep learning framework (adapted S3D architecture) applied to 2D echocardiography (transthoracic echocardiography [TTE] or handheld cardiac ultrasound [HCU]) to automatically estimate left ventricular ejection fraction (LVEF), patient age, and patient sex from any valid view containing the left ventricle.

Comparator

Clinically-calculated LVEF by expert echocardiographers, standard TTE vs HCU, and expert vs novice HCU image acquisition.

Outcome

Model performance for LVEF estimation (measured by r2, RMSE, MAE) and classification of reduced vs normal LVEF (measured by AUC).surrogate

A view-flexible deep learning model can accurately estimate LVEF, age, and sex from both standard TTE and handheld cardiac ultrasound, maintaining strong performance even when images are acquired by novice users.

Main Result

Effect estimate: AUC 0.981

Limitations

Training cohort lacked diversity in terms of race and ethnicity
Handheld cardiac ultrasound data were collected in controlled environments
Novice users utilized real-time AI guidance software during image acquisition, which may not reflect typical point-of-care scenarios
Model may rely in part on non-physiological features such as noise or artefacts to make decisions

Abstract

Echocardiography traditionally requires experienced operators to select and interpret clips from specific viewing angles. Clinical decision-making is therefore limited for handheld cardiac ultrasound (HCU), which is often collected by novice users. In this study, we developed a view-flexible deep learning framework to estimate left ventricular ejection fraction (LVEF), patient age, and patient sex from any of several views containing the left ventricle. Model performance was: (1) consistently strong across retrospective transthoracic echocardiography (TTE) datasets; (2) comparable between prospective HCU versus TTE (625 patients; LVEF r2 0.80 vs. 0.86, LVEF > or ≤40% AUC 0.981 vs. 0.993, age r2 0.85 vs. 0.87, sex classification AUC 0.985 vs. 0.996); (3) comparable between prospective HCU data collected by experts versus novice users (100 patients; LVEF r2 0.77 vs. 0.64, LVEF AUC 0.983 vs. 0.968). This approach may broaden the clinical utility of echocardiography by lessening the need for user expertise in image acquisition.

Demander à l'IA

Bookmark

View Full Paper