What is the clinical evidence from this study?

Study design: Observational. Population: Left ventricular systolic dysfunction (n=191941). Intervention: Personalized ECG-based convolutional neural network vs. Generalized ECG-based convolutional neural network. Primary outcome: Mean absolute error (MAE) of LVEF estimation (p=<0.0001).

What does this research mean for the field?

Artificial intelligence models utilizing electrocardiogram (ECG) data can accurately estimate left ventricular ejection fraction and screen for systolic dysfunction, with personalized models achieving a mean absolute error as low as 5.98%. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This research aims to develop AI models for estimating left ventricular ejection fraction using ECG data.

April 10, 2026Open Access

Personalized artificial intelligence based left ventricular ejection fraction and systolic dysfunction assessment

Q: What are the key findings of this study?

A personalized artificial intelligence model using electrocardiogram data estimated left ventricular ejection fraction with a mean absolute error of 5.98%, significantly outperforming a generalized model.

Q: What does this research mean for the field?

Artificial intelligence models utilizing electrocardiogram (ECG) data can accurately estimate left ventricular ejection fraction and screen for systolic dysfunction, with personalized models achieving a mean absolute error as low as 5.98%. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Q: What question did this study set out to answer?

This research aims to develop AI models for estimating left ventricular ejection fraction using ECG data.

Key Result

A personalized artificial intelligence model using electrocardiogram data estimated left ventricular ejection fraction with a mean absolute error of 5.98%, significantly outperforming a generalized model.

Key Points

This research aims to develop AI models for estimating left ventricular ejection fraction using ECG data.
Evaluated 191,941 patients with 236,623 ECG/TTE pairs.
Developed convolutional and probabilistic neural networks.
Used ECG data alone and combined with structured features for analysis.
ECG-only model achieved MAE of 7.71% and RMSE of 10.36%.
Hybrid model achieved MAE of 7.84% and RMSE of 10.52%.
Personalized models improved accuracy with MAE of 5.98% (ECG-only) and 6.75% (hybrid).
Identified LV systolic dysfunction with AUC of 0.88, sensitivity of 0.92, and negative predictive value of 0.98.

Study Design

Type

Observational (n=191,941)

Structured PICO

Can artificial intelligence models using ECG data accurately estimate LVEF and identify LV systolic dysfunction compared to TTE?

Population

191,941 patients, comprising 236,623 ECG/TTE pairs

Intervention

Convolutional and probabilistic neural network models using ECG data alone or with structured features

Comparator

Transthoracic echocardiography (TTE)

Outcome

Estimation of LVEF and identification of LV systolic dysfunction (LVEF ≤ 40%)surrogate

AI-based analysis of standard ECGs can accurately estimate LVEF and screen for LV systolic dysfunction, offering an accessible alternative to echocardiography.

Main Result

Absolute Event Rate: 5.98% vs 8.98%

p-value: p=<0.0001

Limitations

Higher mean absolute error in extreme LVEF ranges (<25% and >75%) due to limited sample sizes.
Low positive predictive value (24%) for screening left ventricular systolic dysfunction, which could lead to a substantial burden of confirmatory echocardiograms.
Slightly higher error during external validation, potentially due to imaging modality shift or population demographic differences.
Simple early-fusion strategy for the hybrid model may not optimally capture cross-modal interactions.

Abstract

Left-ventricular (LV) ejection fraction (LVEF) is a fundamental measure of cardiac function, typically assessed with resource-intensive imaging techniques, such as transthoracic echocardiography (TTE). We evaluated the electrocardiogram (ECG) as an alternative, easily accessible data to estimate LVEF in a large cohort of 191,941 patients, comprising 236,623 ECG/TTE pairs. Using either the ECG data alone or with structured features, we developed convolutional and probabilistic neural network models to estimate LVEF and quantify its uncertainty. The ECG-only model achieved a mean-absolute-error (MAE) of 7.71% and a root-mean-square-error (RMSE) of 10.36%, while the hybrid model achieved an MAE of 7.84% and an RMSE of 10.52%. Personalized models significantly improved performance, achieving MAEs of 5.98% (ECG-only) and 6.75% (hybrid). LV systolic dysfunction (LVEF ≤ 40%) was identified with an AUC of 0.88, sensitivity of 0.92 and negative predictive value of 0.98. The presented models demonstrated excellent performance in estimating LVEF and screening of LV systolic dysfunction.

Bookmark

View Full Paper