High-performance AI expands electrocardiogram analysis across medical tasks

by Doris Heidegger, Medizinische Universität Innsbruck

edited by Lisa Lock, reviewed by Robert Egan

Lisa Lock

Scientific Editor

Meet our editorial team
Behind our editorial process

Robert Egan

Senior Editor

Meet our editorial team
Behind our editorial process Editors' notes

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

fact-checked

peer-reviewed publication

proofread

The GIST Add as preferred source

Conventional AI models in medicine are usually trained for a single, narrowly defined task, such as detecting a specific cardiac arrhythmia. So-called foundation models take a different approach: Much like a language model such as ChatGPT develops a comprehensive understanding of text, an ECG foundation model learns to "understand" the heart signal in all its diversity. It can then apply that knowledge to many medical questions—including those for which only limited specific training data is available. This is a decisive advantage, especially in medicine.

AI unlocks new potential in ECGs

The developments of the "Digital Medicine in Cardiology" research group, led by Clemens Dlaska at the University Clinic for Internal Medicine III (Cardiology and Angiology), are designed for this purpose.

"Digital patient information plays a crucial role in the prevention, early detection, diagnosis and treatment of diseases, particularly in cardiovascular medicine. Physicians rely on a wide variety of data, such as imaging data and electrocardiograms (ECG), to inform their decision-making. Our goal is to harness the potential of this data from clinical practice using state-of-the-art AI methods, for example, to predict and detect heart attacks, cardiac arrhythmias or sudden cardiac death early," Dlaska reports.

He and his team—first authors Riccardo Lunelli and Angus Nicolson, along with Samuel Martin Pröll—developed the xECG foundation model, drawing on the clinical expertise of cardiologists Axel Bauer and Sebastian Reinstadler. The Innsbruck model combines the xLSTM architecture, recently proposed by a team led by Austrian AI pioneer Sepp Hochreiter at JKU Linz, with a training method from computer vision adapted and applied for the first time to time-series data such as ECGs.

It is among the most powerful and versatile foundation models for ECG data worldwide. In addition, using a specially developed evaluation system, the Innsbruck team has, for the first time, established clear scientific criteria for the quality of ECG foundation models—an important requirement for performance comparisons. Their research is published in the journal npj Digital Medicine.

"Our xECG model can also efficiently process very long signals, such as nighttime recordings for sleep apnea diagnosis, an area where many other models reach their limits. The computational cost of our system scales linearly with the signal length, which is what makes it so efficient, especially with very long ECG signals. By handling a broad spectrum of conceptually diverse tasks—such as classification, regression and survival prediction—it currently outperforms other models," Dlaska says. The Innsbruck foundation model was trained using approximately 8 million ECGs from about 1.7 million patients.

A new benchmark, new quality

The team also provides evidence of xECG's leading role in this field. "Our goal wasn't just to create a single good model. Above all, we wanted to bring clarity and a systematic approach to this field of research, and with BenchECG, we've also created a reliable and freely accessible evaluation framework that the entire research community can now build upon," Dlaska says.

Several research groups worldwide have developed models in recent years and referred to them as "ECG foundation models." Until now, however, there has been no standardized, verifiable definition of their performance capabilities.

With BenchECG, the Innsbruck team has defined three key requirements. First, an efficient and comprehensive foundation model must be able to handle a broad spectrum of conceptually diverse tasks, ranging from classification and the detection of subtle signal features to the prediction of survival probabilities. Second, it must be able to handle different types of ECG recordings, ranging from the standard clinical 10-second 12-lead ECG to long-term ECGs and smartwatch data. Third, it must function reliably across diverse patient groups, from healthy individuals to high-risk patients.

"Using publicly available, highly diverse ECG datasets (approximately 1.7 million ECGs from around 400,000 patients), we systematically measured how well the model's representations transfer across tasks under controlled benchmark conditions. Most of the ECG foundation models we evaluated only partially passed this test; while they performed very well in one area—such as classification—they showed significant weaknesses in other task domains," Dlaska says.

The current study is part of a larger research program at the Department of Cardiology in Innsbruck. "We can cover the entire spectrum from fundamental AI development through concrete applications to clinical trials. The first clinical applications of the model are already in preparation," say Dlaska and Clinic Director Bauer.

Publication details

Riccardo Lunelli et al, BenchECG and xECG: a benchmark and baseline for ECG foundation models, npj Digital Medicine (2026). DOI: 10.1038/s41746-026-03196-y

Journal information: npj Digital Medicine

Key medical concepts

Electrocardiograms

Clinical categories

Cardiology Provided by Medizinische Universität Innsbruck Who's behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021. Full profile →

Robert Egan

Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →

Citation: High-performance AI expands electrocardiogram analysis across medical tasks (2026, September 29) retrieved 29 September 2026 from https://medicalxpress.com/news/2026-09-high-ai-electrocardiogram-analysis-medical.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.