A self-supervised electrocardiogram foundation model for empowering cardiovascular disease prediction and genetic factor discovery

2026-04-27 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Artificial Intelligence & Machine Learning, Medical Specialties & Subspecialties, Medical Devices & Health Technology · Depth: Expert, short

Summary

Researchers developed ECG-LFM, a self-supervised Electrocardiogram Large-scale Foundation Model, to improve cardiovascular disease (CVD) prediction and genetic factor discovery. The model was pre-trained on over ten million 12-lead ECGs from multiple datasets, integrating contrastive learning with masked language modeling to capture both global and fine-grained ECG patterns. When fine-tuned for eight CVD types, ECG-LFM achieved an average AUROC of 0.930 across multiple datasets, outperforming existing methods. The model's derived features (EDFs) represent known CVD biomarkers, demonstrating high interpretability. Furthermore, applying EDFs in genome-wide association studies identified 24 significant single nucleotide polymorphisms (SNPs) associated with ECG, including 8 novel findings, with P-value < 5×10-8 and LD r2 < 0.01. Mendelian randomization also indicated causal relationships between 2 CVDs and 4 EDFs.

Key takeaway

For AI Scientists and Machine Learning Engineers developing diagnostic tools, ECG-LFM demonstrates a robust approach to improving CVD prediction and interpretability. Your teams should consider self-supervised foundation models pre-trained on large-scale medical data to enhance model performance and enable novel genetic insights. This method can lead to more accurate diagnoses and a deeper understanding of disease mechanisms.

Key insights

A self-supervised ECG foundation model improves CVD prediction and uncovers novel genetic associations.

Principles

• Self-supervision enhances ECG representation.
• Integrate global and fine-grained pattern capture.

Method

ECG-LFM uses contrastive learning and masked language modeling for self-supervised pre-training on millions of 12-lead ECGs, then fine-tunes for CVD prediction and applies derived features to GWAS and Mendelian randomization.

In practice

• Use foundation models for medical signal analysis.
• Apply EDFs for genetic association studies.

Topics

• ECG Foundation Model
• Cardiovascular Disease Prediction
• Self-supervised Learning
• Genome-Wide Association Study
• Genetic Factor Discovery

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Masked and Predictive Self-Supervised Foundation Models for 3D Brain MRI — Self-supervised objective design for medical foundation models must align with downstream task structure for optimal performance.
• ECG Signal Classification Using Image-Based CNN Deep Learning — ECG signals can be effectively classified for cardiac abnormalities by transforming them into "2-D scalogram images" for CNN deep learning.
• A spectral audit framework reveals task-dependent aperiodic reliance across EEG and ECG deep learning — Deep learning models on physiological data often rely on aperiodic signals, impacting task performance.
• Context-Aware Hierarchical Bayesian Modeling of IVF Laboratory Environmental Conditions — Context-aware temporal features from IVF lab environments, modeled hierarchically, significantly improve pregnancy rate prediction and are transferable.
• HAPI-EP: Towards Hybrid, Adaptive, and Predictive Digital Twins of Cardiac Electrophysiology — HAPI-EP builds identifiable, predictive digital twins by integrating mechanistic and neural models with meta-learning for rapid adaptation.
• ECG-Lens: Benchmarking ML & DL Models on PTB-XL Dataset — Deep learning models, especially complex CNNs, significantly outperform traditional ML for raw 12-lead ECG classification.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.