Valvular & structural heart diseaseDiagnostic or prognostic study
Deep Learning for Echo Analysis, Tracking, and Evaluation of Mitral Regurgitation (DELINEATE-MR)
- Diagnostic accuracy study
- internal and external test sets
Authors' conclusion
This end-to-end DL system can intake entire echocardiogram studies to accurately classify MR severity and may be useful in helping clinicians refine MR assessments.
Study context
- Population
- Transthoracic echocardiograms assessed for mitral regurgitation
- Participants
- 61,689 TTEs
- Comparison
- Internal test vs External test
- Internal testn = 8,987
- External testn = 8,208
- Outcome
- Agreement with reading cardiologist on 4-step MR severity (none/trace, mild, moderate, severe); weighted κ; AUC for moderate or greater and severe MR
Performance
As reported in the abstract
MR severity classification (4-step) vs reading cardiologist
| Test or model | Exact accuracy | κ | AUC for moderate or greater mitral regurgitation (MR) |
|---|---|---|---|
| Internal test | 82% | 0.84 | 0.98 |
| External test | 79% | 0.80 | 0.98 |
Further findings
Misclassification disagreements between none/trace and mild MR
- Internal test set
- 63%
- External test set
- 66%
Accuracy by TTE views
- Multiple TTE views
- 82%
- Apical 4-chamber views only
- 80%
Primary vs secondary MR and eccentric MR (subset analyses)
- Accurate for both primary and secondary MR; slightly lower performance in eccentric MR
6-step MR classification
- Exact accuracy, internal test set
- 80%
- Exact accuracy, external test set
- 76%
- ±1 step agreement, internal test set
- 99%
- ±1 step agreement, external test set
- 98%
Original publication
Deep Learning for Echo Analysis, Tracking, and Evaluation of Mitral Regurgitation (DELINEATE-MR). Circulation. Published 17 June 2024. doi:10.1161/circulationaha.124.068996
Summarised from the published abstract; every number is checked against it