Artificial Intelligence Applications in Echocardiography

Artificial intelligence (AI) is transforming echocardiography by automating image interpretation, enhancing measurement precision and streamlining clinical workflows. Deep learning algorithms now perform tasks that traditionally relied on expert sonographers, including view classification, chamber segmentation and quantification of left ventricular ejection fraction. Multimodal approaches that combine ultrasound images with text reports enable rapid retrieval of prior studies and preliminary diagnostic suggestions. These technologies reduce inter-operator variability, improve reproducibility of functional parameters and accelerate decision-making in acute and outpatient settings. Recent advances in foundation models have further expanded capabilities, allowing a single AI system to tackle multiple interpretation tasks without task-specific retraining. By lowering the barrier to high-quality cardiac imaging analysis, AI holds promise for standardising care across diverse healthcare environments, augmenting clinician expertise, and extending sophisticated cardiac assessment to under-resourced regions.

Research from Nature Portfolio

Recent trials demonstrate that AI-guided workflows for assessing left ventricular ejection fraction are non-inferior to experienced sonographers. In a large blinded randomised study, initial AI estimates yielded fewer substantial adjustments by cardiologists, reduced interpretation time and achieved parity with human assessment of ejection fraction. A vision–language foundation model trained on over a million echocardiographic videos and corresponding expert text has shown strong performance across diverse tasks, including quantitative ejection fraction prediction, detection of intracardiac devices and long-context patient matching for sequential imaging. This model’s ability to retrieve relevant text reports and identify clinical transitions exemplifies the potential of foundation architectures in cardiovascular imaging. A formal validation of a fully automated deep learning workflow across 23 echocardiographic parameters reported disagreement metrics lower than those among core-lab readers, indicating that automated measurement of volumes, Doppler indices and ejection fraction can match or exceed human reproducibility while improving efficiency and reducing costs.

Research from all publishers

Early demonstrations of convolutional neural networks in echocardiography established accurate classification of standard views, achieving over 97 % accuracy across multiple video orientations and outperforming board-certified echocardiographers on single frames. Subsequent work with large annotated datasets produced models capable of identifying cardiac structures, estimating volumes and ejection fraction, and even predicting systemic phenotypes such as age or sex from ultrasound images. These models employ attention mechanisms to highlight clinically relevant anatomical features, paving the way for preliminary interpretation in settings lacking specialists. More recent studies in cardiology journals have applied deep learning to high-throughput phenotyping of left ventricular hypertrophy, accurately measuring wall thickness and distinguishing causes such as hypertrophic cardiomyopathy or amyloidosis across domestic and international cohorts. The fully automated pipelines demonstrate strong external validity and offer a foundation for precision diagnostics in cardiac disease.

Artificial Intelligence Applications in Echocardiography publication trend

The graph below shows the total number of articles in artificial intelligence applications in echocardiography across all publications each year (not limited to Nature Index journals).

Technical terms

Convolutional neural network (CNN): A deep learning architecture that applies convolutional filters to extract hierarchical image features.

Ejection fraction: The proportion of blood ejected from the left ventricle during systole, expressed as a percentage.

Foundation model: A large AI model trained on broad datasets that can be fine-tuned or applied to diverse downstream tasks without task-specific training.

Global longitudinal strain: A measure of myocardial deformation along the long axis of the left ventricle, reflecting contractile function.

View classification: Automated identification of echocardiographic acquisition angles or standard imaging planes within an ultrasound study.

References

1. Blinded, randomized trial of sonographer versus AI cardiac function assessment. Nature (2023).
2. Vision–language foundation model for echocardiogram interpretation. Nature Medicine (2024).
3. Fast and accurate view classification of echocardiograms using deep learning. npj Digital Medicine (2018).
4. High-Throughput Precision Phenotyping of Left Ventricular Hypertrophy With Cardiovascular Deep Learning. JAMA Cardiology (2022).
5. A formal validation of a deep learning-based automated workflow for the interpretation of the echocardiogram. Nature Communications (2022).