Valvular & structural heart diseaseDiagnostic or prognostic study
Automated Deep Learning Phenotyping of Tricuspid Regurgitation in Echocardiography
- Diagnostic accuracy study
- temporally distinct test set
- geographically distinct external cohort
Authors' conclusion
In this study, an automated pipeline was developed to identify clinically significant TR with excellent performance.
Study context
- Population
- Color Doppler echocardiogram studies; development at Cedars-Sinai Medical Center (CSMC) 2011–2021 (47,312 studies, 31,708 patients), chosen for parity across tricuspid regurgitation (TR) severity classestest studies per cohort
- Participants
- 2,462 vs 5,549
- Comparison
- CSMC test vs Stanford Healthcare (SHC)
- CSMC testn = 2,462
- SHCn = 5,549
- Outcome
- AUC, sensitivity and specificity for identifying A4C videos with color Doppler across the tricuspid valve, and AUC for identifying moderate to severe or severe TR
Performance
As reported in the abstract
| Test or model | View classifier AUC | Studies with ≥1 A4C color Doppler video identified | View classifier sensitivity | View classifier specificity | Moderate or severe TR AUC | Severe TR AUC | Studies with ≥1 TR color Doppler video identified |
|---|---|---|---|---|---|---|---|
| CSMC test | 1.000 (95% CI, 0.999-1.000) | 2410 of 2462 | 0.975 (95% CI, 0.968-0.982) | 1.000 (95% CI, 1.000-1.000) | 0.928 (95% CI, 0.913-0.943) | 0.956 (95% CI, 0.940-0.969) | |
| SHC | 0.999 (95% CI, 0.998-0.999) | 0.949 (95% CI, 0.944-0.955) | 0.999 (95% CI, 0.999-0.999) | 0.951 (95% CI, 0.938-0.962) | 0.980 (95% CI, 0.966-0.988) | 5268 of the 5549 |
Original publication
Automated Deep Learning Phenotyping of Tricuspid Regurgitation in Echocardiography. JAMA Cardiol. Published 1 June 2025. doi:10.1001/jamacardio.2025.0498
Summarised from the published abstract; every number is checked against it