Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.
Report-guided 12-lead ECG FM; macro-AUPRC 0.624 vs 0.615 (DeepECG-SSL) on 45 internal diagnoses, ties externally, strong label efficiency. Paper: the CardioLoom paper.
Signals
ECG, Text
License
MIT (code)
Year
2026
Architecture
ECG encoder initialised from DeepECG-SSL trained against MedCPT report embeddings; report-shared positives
Pretraining data
2,781,289 ECG-report pairs, 1,298,359 patients (Zhongshan; not released)
Routine electrocardiogram (ECG) reports offer scalable supervision for cardiovascular representation learning. In this retrospective study, we pretrained CardioLoom, a report-guided ECG foundation model, on 2,781,289 ECG-report pairs from 1,298,359 patients. Downstream development used a separate cohort of 721,317 ECGs from 233,798 patients. Evaluation covered 89 targets, with external testing in nine cohorts. Across 45 internal ECG diagnoses, macro area under the precision-recall curve (AUPRC) was 0.624 versus 0.615 for DeepECG-SSL. Across 203 external ECG evaluations, corresponding means were 0.425 and 0.424. Using labels from 10% of training patients, CardioLoom achieved macro-AUPRC 0.559, exceeding supervised training with all labels (0.549). For heart failure with reduced ejection fraction, AUPRCs were 0.690, 0.641 and 0.677 in Zhongshan, Shiyuan and BIDMC. Cardiac amyloidosis screening yielded 4.5% positive predictive value using a threshold targeting 99% specificity in validation. Report-guided pretraining supported label-efficient ECG diagnosis, with variable transfer across echocardiographic phenotypes.
Verbatim fromarXiv· arXiv metadata (abstracts are released under CC0)
Key results
0.624
AUPRC
macro-AUPRC 0.624 vs 0.615 (DeepECG-SSL) on 45 internal diagnoses, ties externally, strong label efficiency
How it works
Pretraining data
2,781,289 ECG-report pairs, 1,298,359 patients (Zhongshan; not released)
Architecture
ECG encoder initialised from DeepECG-SSL trained against MedCPT report embeddings; report-shared positives
Inputs
ECG, Text
Approach
ecg fm · report guided · label efficient · no weights
Cite the publication recorded for this model. Check it against the publisher before submitting.
Yu, Z., Tao, Y., Huo, J., et al. (2026). CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping. arXiv:2605.25446. https://arxiv.org/abs/2605.25446
Yu, Ziqing, Yuhui Tao, and Jiayu Huo, et al. 2026. "CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping." arXiv:2605.25446. https://arxiv.org/abs/2605.25446.
Yu, Z., Tao, Y. and Huo, J. et al. (2026) 'CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping', arXiv:2605.25446. Available at: https://arxiv.org/abs/2605.25446
Yu, Ziqing, et al. "CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping." arXiv:2605.25446, 2026. https://arxiv.org/abs/2605.25446.
Z. Yu et al., "CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping," arXiv:2605.25446, 2026, [Online]. Available: https://arxiv.org/abs/2605.25446.
@misc{yu2026cardioloom,
title = {{CardioLoom: A Report-Guided Electrocardiogram Foundation Model for Cross-Domain Cardiovascular Phenotyping}},
author = {Yu, Ziqing and Tao, Yuhui and Huo, Jiayu and others},
year = {2026},
howpublished = {arXiv:2605.25446},
eprint = {2605.25446},
archivePrefix = {arXiv}
}
History
Added to the ChatPPG research hub
Newsletter
Get new PPG datasets, papers and models
Get alerts when CardioLoom (report-guided ECG foundation model) is benchmarked on new datasets, plus the weekly digest.
New datasets, papers and models
Practical guides from the hub
Unsubscribe anytime
The waveform is an illustration of a PPG pulse, not a recording.