Dear members of the ESC Working Group on e-Cardiology,
We are pleased to share a new selection of recent studies highlighting how artificial intelligence continues to unlock the remarkable diagnostic and prognostic potential embedded within the electrocardiogram (ECG). Long regarded as one of the most accessible tools in cardiovascular medicine, the ECG is increasingly proving to contain far richer information than traditionally appreciated.
The papers featured in this issue illustrate how advanced analytical approaches can extract subtle patterns from ECG signals to predict a wide spectrum of clinically relevant conditions, from structural cardiac remodelling and diastolic dysfunction to perioperative risk stratification and postoperative outcomes. Together, these contributions reinforce a growing concept in digital cardiology: that a simple, routinely acquired signal may hold deep physiological insights when analysed through modern computational methods.
We hope this selection will provide stimulating insights for all members interested in the evolving intersection between cardiovascular medicine, digital technologies, and artificial intelligence.
In this first study by Choi et al., published in European Heart Journal – Digital Health (2026), entitled Artificial intelligence-enhanced ECG score for perioperative risk assessment in non-cardiac surgery, the authors evaluated an artificial intelligence–enhanced ECG score (QCG-Critical) for perioperative risk assessment in patients undergoing non-cardiac surgery. In a cohort of more than 46,000 procedures, the AI-ECG model accurately predicted 30-day postoperative mortality and other adverse outcomes. Its predictive performance was superior to traditional risk stratification tools such as the ESC surgical risk category and the Revised Cardiac Risk Index. These findings suggest that a single preoperative ECG analysed through artificial intelligence may provide rapid and clinically meaningful prognostic information for perioperative risk stratification.
In a second study by O’Sullivan et al., published in JACC: Advances (2026), entitled Artificial Intelligence–Enabled ECG for Diastolic Dysfunction in Congenital Heart Disease: A Novel Risk Stratification Tool, the authors investigated the use of an AI-enabled ECG model to detect and grade diastolic dysfunction in adults with congenital heart disease. Analysing more than 6,700 patients from the Mayo Clinic ACHD registry, the AI-ECG algorithm demonstrated significant correlations with echocardiographic parameters, invasive haemodynamic measurements, and biomarkers of cardiac dysfunction. Importantly, higher AI-derived diastolic grades were independently associated with increased mortality. These findings highlight the potential of AI-enhanced ECG analysis as a scalable and non-invasive tool for risk stratification in this complex patient population.
In the third study by Naderi et al., published in Europace (2026), entitled Deep learning to predict left ventricular hypertrophy from the electrocardiogram, the authors developed a deep learning model capable of identifying left ventricular hypertrophy directly from standard 12-lead ECG recordings. Using nearly 49,000 participants from the UK Biobank with cardiac magnetic resonance as the reference standard, the model achieved excellent diagnostic performance and outperformed conventional ECG criteria. External validation in an independent cohort confirmed the feasibility of this approach, supporting the concept that advanced computational analysis of ECG signals may enable scalable screening strategies for structural heart disease.
Finally, in this month’s commented article, we explore a recent study by Nolin-Lapalme et al., commented in detail by Panteleimon Pantelidis, that marks a paradigm shift toward open-source foundation models in electrocardiography. Trained on over 1 million ECGs and rigorously validated across 11 diverse international cohorts, the authors demonstrate that self-supervised, foundational ECG models can act as highly customizable scaffolds. While these models match traditional supervised AI in standard diagnostic interpretation, they significantly outperform them in extracting complex digital biomarkers, such as 5-year atrial fibrillation risk and long QT syndrome genotype, especially in data-scarce environments. By openly releasing their model weights and preprocessing tools for local, plug-and-play fine-tuning, this approach promises to democratize AI deployment across smaller clinical settings, even as critical challenges in privacy, explainability, and real-world clinical integration remain.
Taken together, the first three studies highlight the growing clinical potential of AI-enhanced ECG analysis, showing how advanced algorithms can extract prognostic and diagnostic insights from a simple, widely available signal. The commented article further expands this perspective by introducing open-source ECG foundation models trained on over one million recordings, offering a scalable framework for future ECG-based discovery and clinical deployment.
On behalf of the ESC Working Group on e-Cardiology,
Raffaele De Lucia, MD, FESC
Chairperson Elect ESC WG on e-Cardiology 2024-2026
