Domain-Adapted Fine-Tuning of ECG Foundation Models for Multi-Label Structural Heart Disease Screening

2026-04-25

AI Summary

This study addresses the high cost and limited accessibility of transthoracic echocardiography for structural heart disease (SHD) screening by proposing an efficient electrocardiogram (ECG)-based preliminary screening approach. The method leverages self-supervised domain adaptation of an open-source ECG foundation model on target-domain ECG data, enhanced with parameter-efficient LoRA fine-tuning and a late fusion strategy to enable multi-label detection of six SHD categories. This framework significantly reduces adaptation costs by eliminating the need for training from scratch or conventional feature engineering, while maintaining strong performance: the best model achieves a macro AUROC of 0.8509 and AUPRC of 0.4297; under a parameter-efficient configuration, it attains an AUROC of 0.8501 and a macro F1-score of 0.3691 at a fixed threshold, offering a practical and effective solution for ECG-assisted echocardiographic triage.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Learning to Search
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

Transthoracic echocardiography is the reference standard for confirming structural heart disease (SHD), but first-line screening is limited by cost, workflow burden, and specialist availability. We evaluated whether open pretrained electrocardiogram (ECG) foundation models can support echo-confirmed multi-label SHD detection using the public EchoNext Mini-Model benchmark. Six echocardiography-derived abnormalities were targeted: reduced left ventricular ejection fraction, increased left ventricular wall thickness, aortic stenosis, mitral regurgitation, tricuspid regurgitation, and right ventricular systolic dysfunction. Under a common pipeline, we compared engineered ECG features with gradient boosting, end-to-end waveform learning from scratch, and transfer from open ECG foundation models. We then applied in-domain self-supervised adaptation of an ECG foundation model (ECG-FM) on EchoNext waveforms followed by selective supervised fine-tuning, and evaluated trade-offs between discrimination and adaptation cost. Adapted ECG-FM models achieved the best overall performance: peak macro-AUROC 0.8509 and macro-AUPRC 0.4297, while a parameter-efficient operating point preserved AUROC (0.8501) and attained the highest fixed-threshold macro-F1 0.3691. Late fusion with covariates did not improve threshold-independent discrimination, and evaluated LoRA, alternative backbones, and mixture-of-foundations strategies did not surpass the best adapted single-backbone models. These results indicate that for ECG-based case finding and echocardiography triage, combining target-domain self-supervised adaptation with selective supervised updating of a pretrained ECG backbone is the most effective transfer strategy.

Problem

• structural heart disease
• ECG foundation models
• multi-label screening
• domain adaptation
• echocardiography triage

Innovation

• domain-adapted fine-tuning
• ECG foundation model
• self-supervised adaptation
• multi-label SHD screening
• parameter-efficient transfer learning