Arrhythmias & electrophysiology
QTcNet: a deep learning model for direct heart rate corrected QT interval estimation
Authors' conclusion
QTcNet, trained on large-scale algorithmically labelled data, consistently outperformed conventional algorithms across three independent, external validation datasets.
Study context
- Population
- ECGs with expert QTc measurements (PTB Diagnostic ECG Database, QTcMS, ECGRDVQ)trained on 120,300 ECGs
- Comparison
- QTcNet vs Conventional algorithms
- Outcome
- Cross-cohort error vs expert QTc (MAE, RMSE)
Methodological detailsdeep learning model development and external validation, 3 independent validation datasets, Trained on algorithm-labelled ECGs (60 150 internal hospital cohort, 60 150 MIMIC-IV) after correction for vendor-specific +15 ms bias
- Deep learning model development and external validation
- 3 independent validation datasets
- Trained on algorithm-labelled ECGs (60 150 internal hospital cohort, 60 150 MIMIC-IV) after correction for vendor-specific +15 ms bias
Performance
As reported in the abstract
Cross-cohort error vs expert QTc
| Test or model | Mean absolute error (MAE) | Root mean square error (RMSE) |
|---|---|---|
| QTcNet | 13.4 ms | 22.1 ms |
| Conventional algorithms | 23.4 ms | 40.1 ms |
Mean absolute error (MAE)
QTcNet13.4 ms
Conventional algorithms23.4 ms
Root mean square error (RMSE)
QTcNet22.1 ms
Conventional algorithms40.1 ms
Further findings
Fine-tuning on cardiologist-annotated ECGs (PTB, n = 449)
- Fine-tuning reduced errors only in PTB dataset; no improvement in cross-cohort performance
Model explainability (Integrated Gradient maps)
- Model concentrated on QRS onset and T wave offset
Original publication
QTcNet: a deep learning model for direct heart rate corrected QT interval estimation. Europace. Published 27 October 2025. doi:10.
Summarised from the published abstract; every number is checked against it