In a multicenter study developing an explainable, flow-aware AI pipeline to jointly grade mitral and tricuspid regurgitation from routine transthoracic echo DICOMs, discrimination for clinically significant (≥moderate) MR/TR was excellent in internal testing (AUROC >0.980; NPV >0.970) but showed modest attenuation in patients with atrial fibrillation or reduced LVEF.

Why It Matters To Your Practice

• MR and TR commonly coexist and are assessed from overlapping echo views, making consistent grading time-intensive and variable across readers and settings.

• An explainable, physiology-aware model with strong rule-out performance could support automated triage—especially where expert echo interpretation is constrained.

• Subgroup attenuation in AF or reduced LVEF highlights where clinician oversight may be most needed before acting on AI outputs.

Clinical Implications

• High NPV suggests a potential role as a screening/rule-out layer to deprioritize studies unlikely to have ≥moderate MR/TR, while escalating likely positives for expedited review.

• Expect slightly lower performance when rhythm irregularity (AF) or reduced systolic function complicates Doppler and hemodynamic interpretation; consider tighter human-in-the-loop thresholds in these strata.

• External validation across two additional institutions supports portability, but modest AUROC drops—especially for severe regurgitation—argue for local calibration and ongoing QA.

Insights

• The pipeline is designed for real-world workflow: automated view selection from routine DICOMs, leaflet pose estimation for morphology, and systolic-phase awareness for Doppler interpretation.

• Joint MR/TR grading may better reflect how clinicians interpret overlapping views and shared hemodynamic context than single-valve models.

• Explainability and physiologic constraints are positioned as adoption enablers, addressing a common barrier to deploying “black-box” echo AI.

The Bottom Line

• This explainable echo AI system maintained excellent internal performance and strong rule-out capability, with modest attenuation in AF or reduced LVEF—exactly the patients where you should plan for closer clinician verification.