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

arXiv cs.LG / 4/28/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The study addresses the bottleneck in structural heart disease (SHD) screening by testing whether open ECG foundation models can detect multiple SHD labels using echo-confirmed data from the EchoNext Mini-Model benchmark.
Researchers evaluated several training approaches (engineered ECG features with gradient boosting, training from scratch, and transfer from ECG foundation models) before introducing a two-stage strategy using in-domain self-supervised adaptation plus selective supervised fine-tuning.
The domain-adapted ECG-FM approach delivered the best overall performance, reaching a peak macro-AUROC of 0.8509 and macro-AUPRC of 0.4297, with strong results under a parameter-efficient operating point as well.
Late-fusion methods with additional covariates and alternative adaptation/modeling variants (e.g., LoRA, different backbones, mixture-of-foundations) did not outperform the best adapted single-backbone configuration.
Overall, the findings suggest that for ECG-based case finding and echocardiography triage, combining target-domain self-supervised adaptation with selective supervised updates is the most effective transfer strategy among those tested.

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.