CARMEN

A cardiorespiratory foundation model pretrained on ten physiological waveforms (ECG, ABP, PPG, CVP, CO₂, AWP, ICP, impedance respiration, respiratory flow, PAP) from perioperative and ICU recordings at Seoul National University Hospital.

Paper: https://arxiv.org/abs/2611.15004

Code, examples and documentation: https://github.com/dlcjfgmlnasa/CARMEN

File	Params	d_model	Layers	Heads
carmen-small.pt	52M	512	16	8
carmen-base.pt	174M	768	24	12
carmen-large.pt	410M	1024	32	16

# pip install "carmen-biosignal[hf] @ git+https://github.com/dlcjfgmlnasa/CARMEN"
from carmen import CARMEN
model = CARMEN.from_pretrained("carmen-base")   # or "carmen-small" / "carmen-large"

Citation

@article{carmen2026,
  title         = {A Cardiorespiratory Foundation Model Pretrained on Ten Physiological Signals for Perioperative and Critical Care},
  author        = {Lee, Cheol-Hui and Kim, Wookyung and Lee, Hyeonhoon and Lee, Garam and Eun, Da-in and Lee, Hyung-Chul},
  journal       = {arXiv preprint arXiv:2611.15004},
  year          = {2026},
  eprint        = {2611.15004},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2611.15004}
}

License

PolyForm Noncommercial License 1.0.0 — free for research, education and other noncommercial use. Commercial use requires a separate license: contact dlcjfgmlnasa28@gmail.com or vital@snu.ac.kr.

Research use only. CARMEN is not a medical device and has not been cleared or approved for clinical use.