Valvular & structural heart diseaseDiagnostic or prognostic study
High-Throughput Deep Learning Detection of Mitral Regurgitation
- Diagnostic accuracy study
- internal and external validation
- multicentre
Authors' conclusion
In this study, a novel automated pipeline for identifying clinically significant MR from full transthoracic echocardiography studies demonstrated excellent performance across large numbers of studies and across multiple institutions.
Study context
- Population
- Transthoracic echocardiograms from Cedars-Sinai Medical Center (development, internal test) and Stanford Healthcare (external test)echocardiograms (2,587,538 videos)
- Participants
- 58,614 echocardiograms
- Comparison
- Automated deep learning pipeline for detection of clinically significant mitral regurgitation (MR) on transthoracic echocardiograms
- Cedars-Sinai test setn = 1,800
- Stanford Healthcare testn = 915
- Outcome
- Detection of clinically significant (moderate or severe) MR
Performance
As reported in the abstract
Detection of MR moderate or greater and severe MR
| Test or model | AUC, MR moderate or greater | AUC, severe MR |
|---|---|---|
| Cedars-Sinai Medical Center test cohort | 0.916 (0.899-0.932) | 0.934 (0.913-0.953) |
| Stanford Healthcare test cohort | 0.951 (0.924-0.973) | 0.969 (0.946-0.987) |
Further findings
| Outcome | Cedars-Sinai test set | Stanford Healthcare test |
|---|---|---|
| View classifier: identification of apical 4-chamber color Doppler videos across the mitral valve · vs manually curated videos | ||
| AUC | 0.998 (0.998-0.999) | Not reported in the abstract |
| Correctly identified | 3452 of 3539 | 1051 of 1055 |
| Sensitivity | 0.975 [0.968-0.982] | 0.996 [0.990-1.000] |
| Specificity | 0.999 [0.999-0.999] | 0.999 [0.999-0.999] |
Original publication
High-Throughput Deep Learning Detection of Mitral Regurgitation. Circulation. Published 12 August 2024. doi:10.1161/circulationaha.124.069047
Summarised from the published abstract; every number is checked against it