Daijiworld Media Network – Mumbai

Mumbai, Sep 2: Scientists at Scripps Research have developed a new artificial intelligence model that could improve the detection and prediction of several heart diseases, particularly in clinical settings where limited labelled medical data or fewer ECG leads are available.

The new AI tool, called ECG-CLIP, was trained using more than 1.7 million electrocardiograms (ECGs) collected from over 540,000 people. The ECGs were paired with clinicians’ notes, allowing the model to learn both the electrical patterns of the heart and relevant clinical information.

The findings were published in The Lancet Digital Health on September 1, 2026.

ECG-CLIP is described as a foundation model, meaning it can learn from large and diverse datasets and subsequently be adapted for multiple medical tasks. Researchers led by Scripps Research scientist Quer tested the model against several existing AI systems to assess its ability to detect diseases, predict future heart conditions and forecast adverse health outcomes.

For disease detection, researchers tested the models for acute myocardial infarction, cardiac amyloidosis and hypertrophic cardiomyopathy using a dataset containing more than 800,000 ECGs.

ECG-CLIP consistently outperformed standard deep-learning and linear models in identifying all three conditions. Researchers also found that it achieved performance comparable to the next-best model trained on the full dataset while requiring about 91% less hand-labelled training data on average.

The model showed particular advantages when only a small number of labelled examples were available. Compared with other foundation models trained on ECG data but without clinicians’ notes, ECG-CLIP performed better when training data were limited, including situations with as few as 10 positive examples of a disease.

Researchers said this could be especially valuable for rare diseases, where obtaining large numbers of accurately labelled ECGs can be difficult.

ECG-CLIP also showed promising results when analysing single-lead ECG recordings for detecting acute myocardial infarction. This could make the technology useful in settings where access to conventional 12-lead ECG systems is limited.

The researchers then tested the model’s ability to predict future heart disease, focusing on atrial fibrillation, an irregular heart rhythm. ECG-CLIP outperformed the other models in predicting future atrial fibrillation from 12-lead ECGs that initially showed normal heart rhythms.

The model also performed best in predicting several adverse health outcomes. These included the likelihood of surviving 30 days after an emergency department visit or surgery, as well as the development of chronic kidney disease and type 2 diabetes within three years.

To improve the interpretability of the AI system, the researchers also generated saliency maps showing which portions of an ECG signal contributed most to the model’s predictions. The researchers said this could help clinicians better understand how the model reaches its conclusions and potentially increase confidence in its use in clinical practice.

The team plans to expand the range of data available to ECG-CLIP and assess its performance in specific clinical environments, including emergency departments. Researchers also hope to determine whether the model can work with different ECG recording systems, including wearable devices, which could eventually support continuous and remote monitoring of heart health.

However, the researchers cautioned that further validation through prospective clinical trials will be necessary before the model can be established for routine use in real-world clinical settings.

The latest work builds on previous AI research from Quer’s laboratory, including an algorithm developed in 2024 that could detect heart attacks and rhythm abnormalities using only three of the standard 12 ECG leads, and a 2023 system designed to identify patients at higher risk of atrial fibrillation using data from a two-week ECG patch.

Researchers said the findings highlight the potential of AI-powered ECG analysis to assist clinicians and cardiologists in detecting cardiovascular disease, particularly when labelled medical data are scarce.