What does this research mean for the field?

A deep learning model based on 12-lead ECGs can accurately detect and discriminate between acute myocardial infarction subtypes, outperforming physician interpretation for STEMI detection. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To train and externally validate a 12-lead electrocardiogram deep-learning model capable of classifying acute myocardial infarction subtypes to guide prompt clinical intervention.

September 15, 2026npj Digital MedicineOpen Access

Training and validation of a 12-lead ECG-based deep-learning model for myocardial infarction subtypes

Q: What are the key findings of this study?

A 12-lead ECG-based deep-learning model outperformed physician interpretation for the detection of ST-segment elevation myocardial infarction (AUROC 0.96 vs 0.89, p<0.001).

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Suspected acute myocardial infarction (n=185357). Intervention: 12-lead ECG-based deep-learning model vs. Physician interpretation. Primary outcome: Detection of ST-segment elevation myocardial infarction (STEMI) measured by AUROC (95% CI 0.94-0.98, p=<0.001).

View Full Paper

Ask AI

Bookmark

Key result

Deep-learning ECG model outperforms physicians in detecting STEMI with an AUROC of ~0.96.

95% CI 0.94-0.98
P<0.001
n=185,357

Why the study?

Accurate detection and differentiation of acute myocardial infarction subtypes from 12-lead ECGs remains challenging, necessitating improved diagnostic tools.

Does a 12-lead ECG-based deep-learning model improve the diagnostic accuracy of detecting acute myocardial infarction subtypes compared to physician interpretation in patients with suspected AMI?

Population

173,396 hospitalized patients for training, 7,591 patients for internal validation, 4,370 patients with suspected AMI for external validation

Comparison

12-lead ECG-based deep-learning model vs physician interpretation for STEMI detection

Design

Observational cohort study with internal and external validation

Authors

TZTobias Zimmermann ISIvo StrebelCross-Cutting Cardiology PLPedro López‐AyalaGeneral Cardiology

Discussion

Member takes

Implication

May enhance STEMI detection over physicians; leaves open prospective validation before clinical adoption.

Key Points

To train and externally validate a 12-lead electrocardiogram deep-learning model capable of classifying acute myocardial infarction subtypes to guide prompt clinical intervention.
A convolutional neural network was trained on 12-lead ECGs from 173,396 hospitalized patients.
The algorithm underwent fine-tuning and internal validation in 7591 patients, followed by external validation in 4370 patients with suspected acute myocardial infarction using prospective cohorts with central diagnostic adjudication.
For ST-segment elevation myocardial infarction, the deep learning model outperformed physician interpretation (AUROC 0.96 [95%-CI 0.94–0.98] vs. 0.89 [0.85–0.93], p < 0.001).
For non-ST-segment elevation myocardial infarction, the model achieved an AUROC of 0.81 [95%-CI 0.79–0.83], with type 1 reaching 0.82 [0.80–0.84] versus type 2 at 0.75 [0.71–0.79].
Occlusion myocardial infarction detection yielded an AUROC of 0.91 [95%-CI 0.89–0.93], demonstrating robust overall calibration across evaluated acute myocardial infarction subtypes.

Study Design

Type

Observational (n=185,357)

Multicenter

Yes

Structured PICO

Does a 12-lead ECG-based deep-learning model improve the diagnostic accuracy of detecting acute myocardial infarction subtypes compared to physician interpretation in patients with suspected AMI?

Population

185,357 hospitalized and emergency department patients with suspected acute myocardial infarction evaluated across training, internal validation, and external validation cohorts.

Exposure

12-lead ECG-based deep-learning model (ResNet architecture) for multi-label prediction of acute myocardial infarction subtypes

Comparator

Physician interpretation of 12-lead ECG (specifically evaluated for STEMI detection)

Outcome

Diagnostic accuracy measured by the area under the receiver operating characteristic curve (AUROC) for detecting acute myocardial infarction subtypes (NSTEMI, STEMI, OMI)

Main Result

Absolute Event Rate: 0.96% vs 0.89%

p-value: p=<0.001

A 12-lead ECG-based deep learning model demonstrated high diagnostic accuracy for detecting acute myocardial infarction subtypes, significantly outperforming physician interpretation for STEMI detection.

Limitations

Potential misclassification of a few patients despite rigorous central adjudication
Underrepresentation of STEMI patients who bypass the emergency department for direct invasive coronary angiography
Initial emergency department ECGs may lack characteristic diagnostic features, potentially underestimating model performance
Differences in AMI prevalence between training and validation cohorts may affect generalizability
Model performed better in younger patients without known coronary artery disease, limiting utility in patients with chronic ischemic changes
Uncertainty regarding whether the chosen ResNet architecture is the optimal choice for ECG data classification
Underrepresentation of STEMI patients who often bypass the ED for direct invasive coronary angiography
Initial ED ECGs may lack characteristic diagnostic features, potentially causing an underestimation of model performance
AMI prevalence differed between training and fine-tuning/validation cohorts
Model performed better in younger patients without known CAD due to fewer chronic ischemic changes
ResNet architecture was chosen but it remains unknown if it is the absolute best architecture for this task

Cite This Study

Zimmermann et al. (2026) conducted an observational in Suspected acute myocardial infarction (n=185,357). 12-lead ECG-based deep-learning model vs. Physician interpretation was evaluated on Detection of ST-segment elevation myocardial infarction (STEMI) measured by AUROC (95% CI 0.94-0.98, p=<0.001). A 12-lead ECG-based deep-learning model outperformed physician interpretation for the detection of ST-segment elevation myocardial infarction (AUROC 0.96 vs 0.89, p<0.001).

synapsesocial.com/papers/6aa9016aeed42882c1fc184f https://doi.org/10.1038/s41746-026-03204-1

Topics

Artificial intelligence in cardiology

View Full Paper

Ask AI

Bookmark

Training and validation of a 12-lead ECG-based deep-learning model for myocardial infarction subtypes

Discussion

Member takes

Implication

Key Points

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Topics

Training and validation of a 12-lead ECG-based deep-learning model for myocardial infarction subtypes

Discussion

Member takes

Implication

Key Points

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Also Consider

Topics

Also Consider