[Electrocardiogram Artificial Intelligence modeling]

This research project aims to develop and implement a robust framework for AI in cardiology. By utilizing raw ECG data, we develop tools that provide physicians with efficient, non-invasive decision support. Beyond initial diagnosis, this project focuses on the safe, longitudinal monitoring of patients tracking how AI-derived health scores change over time.

---

Objective

Our goal is to build accurate and reliable AI models that analyse ECG signals to predict a range of cardiac conditions. These predictions are integrated into a digital dashboard for cardiologists.

Methodology

The project is organized into three pillars:

• Foundation Model Development:
We utilise a “Foundation Model” architecture, which learns the complex patterns of the heart from hundreds of thousands of ECGs. This serves as a base to develop specific detection tasks like cardiac amyloidosis, hypertrophic cardiomyopathy, paroxysmal atrial fibrillation, low ejection fraction and biological age.

• Clinical Validation:
Through the Belgian SHARE network, our cardiac amyloidosis model is being validated to ensure it performs generalizable across all participating hospitals.

• Longitudinal Monitoring:
We investigate the use of advanced statistics to track AI scores over time instead of single snapshot predictions.

Impact and future directions

As the ECG is a low-cost, non-invasive measure, these AI tools enable the identification of cardiac irregularities without invasive procedures. By moving from a single “snapshot” diagnosis to longitudinal tracking, we provide cardiologists with a tool to monitor the trajectory of a patient. We will further investigate whether this approach leads to earlier detection and better detection performance. While current work focuses on cardiac abnormalities, the framework can be expanded to non-cardiac outcomes in the future, bringing AI technology closer to the patient in a safe and responsible way.

General info and contact

Keywords (#): ECG, AI, non-invasive

RADar project research lead: Ir. Louise Vander Heyde

Principal investigator: MD Karl Dujardin

Status: Ongoing

Publications / presentations:

Abstract submitted and accepted for oral presentation at the European Congress of cardiology 2025

L Vander Heyde, K Dujardin, W Anne, M Vanhaverbeke, D Mc Auliffe, N Mertens, P De Jaeger, AI-derived ECG age gap: a novel predictor of mortality after CABG and PCI, European Heart Journal, Volume 46, Issue Supplement_1, November 2025, ehaf784.1811, https://doi.org/10.1093/eurheartj/ehaf784.1811

Abstract submitted and accepted for poster presentation at the European Congress of cardiology 2025

D Mcauliffe, K Dujardin, L Vander Heyde, W Anne, M Vanhaverbeke, P De Jaeger, Novel self-supervised learning outperforms traditional AI in ECG-based transthyretin cardiac amyloidosis detection for earlier diagnosis, European Heart Journal, Volume 46, Issue Supplement_1, November 2025, ehaf784.2316, https://doi.org/10.1093/eurheartj/ehaf784.2316

Paper published:

Louise Vander Heyde, Karl Dujardin, Wim Anné, Maarten Vanhaverbeke, David McAuliffe, Nathalie Mertens, Peter De Jaeger, Artificial intelligence—derived electrocardiographic age gap as a predictor of mortality after coronary revascularization: prognostic value and short-term intra-patient variability, European Heart Journal – Digital Health, Volume 7, Issue 3, April 2026, ztag039, https://doi.org/10.1093/ehjdh/ztag039