Heart failure & cardiomyopathies
Artificial Intelligence–Enabled Electrocardiogram (ECG) Screening for Left Ventricular Systolic Dysfunction (LVSD) in Left Bundle Branch Block (LBBB): Evaluating Model Development and Transfer Learning Approaches
Authors' conclusion
Our findings indicate that a broad AI-ECG model reliably detects LVSD in LBBB patients, and transfer learning offers modest improvements without requiring curated LBBB data sets. Evaluating algorithms in representative clinical populations is essential.
Study context
- Population
- Patients with left bundle branch block (LBBB)ECGs used for external validation · Development ECGs from 4 hospitals 364,845 · External validation ECGs from another hospital 1,334
- Participants
- 1,334 patients
- Index test or model
- 4 AI-ECG models for detecting LVSD in LBBB patients
- Target condition
- Left ventricular systolic dysfunction (LVSD), defined as an ejection fraction ≤40%
Methodological details4 hospitals for development, external validation at another hospital
- 4 hospitals for development
- External validation at another hospital
Performance
As reported in the abstract
External validation
| Test or model | AUROC |
|---|---|
| Transfer learning model | 0.903 (95% CI: 0.887-0.918) |
| General model | 0.899 (95% CI: 0.883-0.915) |
| Automated LBBB model | 0.879 |
| Expert-curated LBBB model | 0.841 |
AUROC
Transfer learning model0.903 (95% CI: 0.887-0.918)
General model0.899 (95% CI: 0.883-0.915)
Automated LBBB model0.879
Expert-curated LBBB model0.841
Further findings
Sensitivity and specificity · External validation
- The general model demonstrated high sensitivity
- The transfer learning model exhibited superior specificity
Original publication
Artificial Intelligence–Enabled ECG Screening for LVSD in LBBB: Evaluating Model Development and Transfer Learning Approaches. JACC Adv. Published 21 August 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →