Heart failure & cardiomyopathiesDiagnostic or prognostic study
Deep Learning-Derived Myocardial Strain
- Diagnostic accuracy study
- external validation
Authors' conclusion
The open-source DLS provides lower variation than human measurements and similar quantitative results.
Study context
- Population
- Echocardiography patients; external dataset and populations with cardiac hypertrophy, incl. advanced cardiac amyloidosis
- Comparison
- Deep learning strain vs conventional 2-dimensional global longitudinal strain (GLS)
- Outcome
- Agreement of DLS with GLS
Performance
As reported in the abstract
Agreement with 2-dimensional GLS (external validation)
- moderate agreement
- ICC 0.56; P = 0.002
- Agreement with 2-dimensional GLS (external validation)
- bias −3.31 % (limits of agreement CI −11.65% to 5.02%)
Further findings
| Outcome | Estimate | P value |
|---|---|---|
| Agreement with 2-dimensional GLS in advanced cardiac amyloidosis — DLS vs 2-dimensional GLS | ICC 0.64 (95% CI 0.53–0.72) | P < 0.001 |
| Agreement with 2-dimensional GLS in advanced cardiac amyloidosis — DLS vs 2-dimensional GLS | bias 0.57 % (limits of agreement CI −4.87% to 6.01%) |
Traditional GLS variability (2 sonographers, 2 vendors)
- ICC between vendors
- 0.29 (95% CI: −0.01 to 0.53, P = 0.03)
- ICC between sonographers
- 0.63 (95% CI: 0.48-0.74, P < 0.001)
Simulated contour variation on speckle tracking (6-pixel shift)
- Mean absolute error
- 3.48%
- Proportional strain difference
- 13.52%
Deep learning strain (DLS) across hypertrophy populations
- DLS differed between populations with cardiac hypertrophy
- P < 0.0001
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
Deep Learning-Derived Myocardial Strain. JACC Cardiovasc Imaging. Published 27 March 2024. doi:10.1016/j.jcmg.2024.01.011
Summarised from the published abstract; every number is checked against it