ECG deep learning ensemble separates Takotsubo syndrome from anterior STEMI and improves reader accuracy
A preprint reports an ensemble of three deep learning models that distinguished Takotsubo syndrome from acute anterior ST-elevation myocardial infarction on 12-lead ECG images with 91.4% accuracy in a 95-person external cohort, and raised 70 healthcare providers' estimated probability of a correct diagnosis from 0.632 unaided to 0.864 with AI assistance.
- Model
- Ensemble of three deep learning models with self-supervised pretraining
- Field
- Medicine
- Date
- 2026-10-19
- Human collaborators
- Eisuke Kagawa, Keigo Dote
What was found
The models were pretrained self-supervised and fine-tuned on ECG images from two hospitals recorded within 14 days of symptom onset, with 832 participants for development and 95 for external validation, healthy controls included. External AUROC was 0.970. A four-phase reader study compared unaided reading, a reference sheet and AI assistance, then randomised readers on a new ECG set to the reference sheet alone, AI predictions, or AI predictions with Grad-CAM heatmaps, which gave a further gain. In exploratory analyses less experienced readers gained most, and cardiologists' unaided advantage reversed under AI assistance, which the authors frame as a utilisation bias hypothesis.
Novelty check
Europe PMC, 2026-10-29: TITLE_ABS:"Takotsubo" AND (TITLE_ABS:"deep learning" OR TITLE_ABS:"artificial intelligence") returned 14 records, including a 2024 JAHA study of AI-enabled ECG for prognosis in Takotsubo and a 2026 study of AI ECG analysis in suspected STEMI; none paired a Takotsubo-versus-STEMI model with a randomised reader study.
Caveats and known objections
A medRxiv preprint, not peer-reviewed, so capped at claimed; version 2 was posted on 2026-10-29. Retrospective, from two Japanese hospitals, with a small external cohort of 95 that includes healthy controls, which makes the task easier than the clinical one of separating two causes of chest pain with ST changes. No funder or vendor is named in the abstract. Readers were scored on ECG sets, not patients, the endpoint is diagnostic accuracy, and the experience-related findings are exploratory. Autonomy is ai-assisted: the model's prediction was shown to readers who made the call.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Claimed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded claimed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). ECG deep learning ensemble separates Takotsubo syndrome from anterior STEMI and improves reader accuracy. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-19-takotsubo-ecg-deep-learningBibTeX
@misc{whataifound-northmedicalcenterhiroshimacityasacitizenshospital-2026-learning,
title = {ECG deep learning ensemble separates Takotsubo syndrome from anterior STEMI and improves reader accuracy},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by North Medical Center Hiroshima City Asa Citizens Hospital. Verification: Claimed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-10-19-takotsubo-ecg-deep-learning}
}