Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
ECG-Based AI Models Enable Cost-Effective Cardiac Function Assessment
Researchers developed AI models that estimate left ventricular ejection fraction from electrocardiograms, offering an accessible alternative to echocardiography.
Background
Left ventricular ejection fraction (LVEF) is typically assessed by echocardiography. Thambiraj and colleagues developed machine learning models to estimate LVEF directly from electrocardiograms (ECGs), which are cheaper and more widely available. The study analyzed 236,623 ECG/echocardiography pairs from 191,941 patients.
Key Findings
- Standard CNN models achieved 7.71% mean absolute error for LVEF estimation from ECG data
- Personalized models improved accuracy to 5.98% error using individual patient longitudinal data
- For detecting systolic dysfunction (LVEF ≤40%): 0.88 AUC, 0.92 sensitivity, 0.98 negative predictive value
- Key ECG features: PR interval, QRS duration, QT interval, and T-wave morphology
- Probabilistic models provided uncertainty estimates with wider intervals at extreme LVEF values
Why It Matters
ECG-based LVEF estimation could serve as an interim triage tool where imaging is unavailable or delayed. Personalized models are valuable in chronic care with repeated patient encounters. This approach enables cost-effective cardiac assessment in resource-limited communities.
Limitations
Models were trained on retrospective single-center data. While external validation using MIMIC-IV was performed, real-world deployment in diverse populations requires further investigation.
Original paper: Personalized artificial intelligence based left ventricular ejection fraction and systolic dysfunction assessment. — NPJ digital medicine. 10.1038/s41746-026-02462-3
