Deep learning for echocardiographic assessment and risk stratification of aortic, mitral, and tricuspid regurgitation: the DELINEATE-regurgitation study.

Summary

Using 71,660 echocardiograms (1.2M color Doppler videos), a multiview deep learning system achieved substantial agreement with cardiologists for AR/MR/TR severity and predicted MR progression with an adjusted HR of 4.1, outperforming single-view approaches.

Key Findings

• Multiview AI achieved weighted kappa 0.81/0.76 (internal/external) for AR, 0.76/0.72 for MR, and 0.73/0.64 for TR, exceeding single-view approaches.
• AI score predicted progression from mild–moderate MR to ≥moderate-severe MR with adjusted HR 4.1 (95% CI 2.5–6.6).
• Leveraged 71,660 TTEs and 1,203,980 color Doppler videos across two centers.

Clinical Implications

AI-assisted echo could harmonize grading of AR/MR/TR, streamline workflows, and identify MR patients at high risk of progression for closer follow-up, timely referral, and optimized timing of intervention.

Limitations

• Retrospective, two-center design; generalizability to other scanners, protocols, and populations needs testing.
• Ground truth relies on clinician labels; potential for labeling variability and bias.

Future Directions

Prospective multicenter trials to assess clinical workflow integration, guideline alignment, and outcome impact; expansion to regurgitation etiology and intervention timing recommendations.