FOUND-AF: Benchmarking ECG Foundation Models for Atrial Fibrillation Detection
Amirhossein Taleshinosrati, Yangyang Wang, Atitaya Phoemsuk, Vahid Abolghasemi, Naser Hossein Motlagh, Sadasivan Puthusserypady, Daniel Teichmann, Abdolrahman Peimankar
Why It Matters
What makes this one worth your time
This work addresses the inconsistency in evaluating ECG models, enabling researchers and clinicians to select the best-performing models for practical applications in AF detection.
FOUND-AF provides a robust benchmarking framework for ECG models in atrial fibrillation detection.
Summary
The paper introduces FOUND-AF, a benchmarking framework for evaluating ECG foundation models specifically for atrial fibrillation detection, using standardized conditions across multiple datasets.
Key contributions
- Development of a leakage-controlled benchmarking framework for ECG models.
- Evaluation of nine foundation models across four heterogeneous datasets under standardized conditions.
- Identification of the ECGFounder model as the top performer in terms of accuracy and efficiency.
Notable insights
- The use of a unified benchmarking framework helps mitigate issues of dataset and preprocessing variability in model evaluation.
- The study highlights the importance of computational efficiency alongside accuracy in clinical applications.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2608.03597v1 Announce Type: new Abstract: Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and is associated with increased risks of stroke, heart failure, and mortality. Recent ECG foundation models offer transferable representations for automated AF detection. However, their relative effectiveness remains unclear because existing studies use different datasets, preprocessing procedures, classifiers, and validation protocols. This study presents FOUND-AF, a unified, leakage-controlled, and deployment-oriented benchmarking framework that evaluates the quality of pretrained ECG representations under identical experimental conditions. Nine publicly available foundation models from five families, including HuBERT-ECG, CLEF, ST-MEM, ECG-JEPA, and ECGFounder, were evaluated across four heterogeneous ECG datasets, namely AFDB, CinC2017, CPSC2021, and LTAFDB. All models were used as frozen feature extractors with standardized preprocessing, model-native resampling, a fixed XGBoost classifier, and recording-level grouped cross-validation. The evaluation included classification metrics, receiver operating characteristic analysis, paired recording-level bootstrap comparisons with Holm correction, embedding-space visualization, and computational efficiency profiling. The ECGFounder model consistently achieved the strongest overall performance across datasets while offering a favorable trade-off between accuracy, model size, inference time, and memory usage. FOUND-AF therefore provides a reproducible framework for selecting ECG foundation models and demonstrates that compact, clinically pretrained encoders can support robust and computationally efficient AF detection across heterogeneous acquisition settings.