ECG Foundation Models Benchmark — Code (FOUND-AF)

Code for benchmarking pretrained ECG foundation models as frozen feature extractors for atrial fibrillation (AF) detection, evaluated across four public PhysioNet ECG datasets.

This is the code-only companion to an ongoing research project. Paper source and results are kept in a separate private repository until publication.

Repository structure

.
├── code/                 Benchmark notebooks (one per dataset)
│   ├── 00_environment_diagnostics.ipynb
│   ├── AFDB.ipynb
│   ├── CINC2017.ipynb
│   ├── CPSC2021.ipynb
│   └── LTAFDB.ipynb
├── data/
│   └── README.md         Dataset sources and preprocessing pipeline
├── requirements.txt
└── LICENSE

What each notebook does

Each dataset notebook:

• clones the original model repositories and downloads pretrained checkpoints (HuBERT-ECG, CLEF, ST-MEM, ECG-JEPA, ECGFounder) from their official sources (GitHub / Hugging Face Hub / Zenodo / Google Drive mirrors),
• loads the dataset's processed windows,
• extracts frozen embeddings from all 9 models,
• runs 5-fold recording-level cross-validation with a fixed-configuration XGBoost classifier,
• writes per-model metrics and efficiency numbers.

Getting started

1. See data/README.md for the four public datasets used, download links, and the exact preprocessing pipeline (10-second windows, 5-second overlap, per-window z-score normalisation).
2. Install dependencies:

   pip install -r requirements.txt

3. Open a notebook in code/ and update the PROJECT_DIR / PARQUET_PATH variables in the first cell to point at your local data location (the notebooks were originally run on Google Colab against a mounted Drive folder).

License

Code is released under the MIT License. The underlying ECG databases and pretrained model checkpoints are third-party resources under their own licenses — see data/README.md.