Heart failure & cardiomyopathiesDiagnostic or prognostic study
Beyond Doppler: Scalable AI Detection of Left Ventricular Outflow Tract (LVOT) Obstruction in HCM
- development and external validation of a multi-view deep learning model
- cross-attention-based video-to-video fusion framework
- 3 standard transthoracic echocardiographic views
- EchoPrime-derived video representations
Authors' conclusion
These results suggest LVOT physiology is encoded in routine 2-dimensional imaging and can be leveraged for clinically relevant gradient classification without Doppler input. The proposed artificial intelligence-guided strategy demonstrates substantial cost savings compared with the screen-all approach. By integrating complementary spatial-temporal information across multiple views, our approach generalizes robustly across populations and may enable real-time decision support, extend LVOT assessment to portable or resource-limited settings, and complement Doppler-based evaluation for longitudinal hypertrophic cardiomyopathy management.
Study context
- Population
- Patients with hypertrophic cardiomyopathy undergoing routine 2-dimensional echocardiographyderivation cohort (tertiary care system in the United States) N=1,833 · internally held-out test set N=275 · Korean external validation cohort N=46
- Comparison
- Multi-view deep learning model using routine 2-dimensional echocardiographic windows to classify LVOT obstruction (>20 mm Hg) without Doppler imaging
Further findings
| Outcome | Single-view baselines | EchoPrime | multi-view fusion (late fusion model) |
|---|---|---|---|
| Performance of multi-view fusion on external cohort · external cohort with significant population-shift | Not reported in the abstract | Not reported in the abstract | area under the receiver operating curve 0.84 |
- Discrimination of single-view baselines
- 0.47-0.70 · external area under the receiver operating curves
- Single-view performance of EchoPrime
- 0.75-0.80 · internal · external 0.79-0.83
Original publication
Summarised from the published abstract; every number is checked against it