ChatPPG.
Paper

Language-supervised electrocardiogram interpretation with vision transformer

Xue Zhou, Tian-Hui Li, Keijiro Nakamura, Hiromasa Hayama, Shing-Hong Liu, Tetsuo Sasano +2 more
Health Information Science and Systems2026ECGCC BY-NC-ND 4.0
Read paper PubMed Cite
Published articlePublication date: Oct 13, 2026Added to atlas: Oct 14, 2026Source
Citations0Crossref referencesChecked Oct 16, 2026
GitHub stars—Not available
HF downloads—Not available

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.

This study investigated whether vision-language learning from paired ECG recordings and summary reports could enable transferable ECG representations without task-specific supervised training. The authors developed ECG-VLM, a contrastive vision-language model consisting of a Transformer-based image encoder and a text encoder, trained on 800,034 paired 10-s ECG recordings and summary reports from the MIMIC-IV-ECG database. Cardiac condition detection was performed using natural-language prompts without condition-specific retraining or fine-tuning.

  • Signals: ECG
  • Result: ECG-VLM achieved diagnostic performance across 14 cardiac conditions.
  • Result: Performance was strongest for rhythm abnormalities (mean AUROC 0.948 internally and 0.921 externally) and lower for morphological conditions.
At a glance
Venue
Health Information Science and Systems (2026-10-13)
Published
2026
DOI
10.1007/s13755-026-00482-7 (opens in a new tab)
arXiv
Not reported
PMID
42876871 (opens in a new tab)
Licence
CC BY-NC-ND 4.0 (opens in a new tab)
Signals
ECG
Form factor
Clinical
Subjects
45,151 recordings
Code
Not reported
Added
Oct 14, 2026

Abstract

Most artificial intelligence models for electrocardiogram (ECG) interpretation rely on task-specific supervised learning, requiring dedicated labeled datasets for each diagnostic task and limiting scalability across diverse cardiac conditions. This study investigated whether vision-language learning from paired ECG recordings and summary reports could enable transferable ECG representations without task-specific supervised training. We developed ECG-VLM, a contrastive vision-language model consisting of a Transformer-based image encoder and a text encoder, trained on 800,034 paired 10-s ECG recordings and summary reports from the MIMIC-IV-ECG database. Cardiac condition detection was performed using natural-language prompts without condition-specific retraining or fine-tuning. Performance was evaluated on an internal test set and externally validated using the LSAS-ECG dataset (45,151 recordings) across 14 cardiac conditions, together with retrieval and subgroup analysis. ECG-VLM achieved diagnostic performance across 14 cardiac conditions, with AUROC ranging from 0.679 to 0.981 on the internal test set and from 0.691 to 0.991 on external validation. Performance was strongest for rhythm abnormalities (mean AUROC 0.948 internally and 0.921 externally) and lower for morphological conditions. Compared with the previously reported signal-based ECG-CLIP framework, ECG-VLM achieved comparable performance for rhythm abnormalities while improving AUPRC for several morphological conditions. Performance remained broadly consistent across external validation and demographic and clinical subgroups. These findings demonstrate the feasibility of vision-language supervision from paired ECG recordings and summary reports for learning transferable ECG representations without task-specific supervised training. This framework provides a basis for future extension of vision-language learning to ECG images encountered in routine clinical practice.

Verbatim fromSemantic Scholar· Abstract via Semantic Scholar; © the authors/publisher

Language-supervised electrocardiogram interpretation with vision transformer: ChatPPG research summary card

How it works

How Method works: Setup → Method → ResultSetup: Cohort (45,151 recordings); Signals (ECG). Method: Method (Transformer). Result: Result (ECG-VLM achieved diagnostic performance across 14 cardiac conditions.); Result (Performance was strongest for rhythm abnormalities (mean AUROC 0.948 internally and 0.921 externally) and lower for morphological conditions.).SetupCohort: 45,151 recordingsCohort45,151 recordingsSignals: ECGSignalsECGMethod×NMethodTransformerResultResult: ECG-VLM achieved diagnostic performance across 14 cardiac conditions.ResultECG-VLM achieved diagnostic performance…Result: Performance was strongest for rhythm abnormalities (mean AUROC 0.948 internally and 0.921 externally) and lower for morphological conditions.ResultPerformance was strongest for rhythm…
Cohort
45,151 recordings

Cite

@article{zhoulanguagesupervisedec2026,
  title = {{Language-supervised electrocardiogram interpretation with vision transformer}},
  author = {Xue Zhou and Tian-Hui Li and Keijiro Nakamura and Hiromasa Hayama and Shing-Hong Liu and Tetsuo Sasano and Wen-Xi Chen and Xin Zhu},
  year = {2026},
  journal = {Health Information Science and Systems (2026-10-13)},
  doi = {10.1007/s13755-026-00482-7}
}

History

  1. Added to the ChatPPG research hub

More on ChatPPG