Automated Classification of Mitral and Tricuspid Regurgitation With Explainability and Real-World Practice Experience

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Mitral regurgitation and tricuspid regurgitation often coexist and are assessed using similar echocardiographic views.
• Existing AI models for valvular heart disease lack explainability and physiological constraints, hindering clinical adoption.
• There is a need for reliable, explainable AI tools in echocardiographic workflows.

Purpose of the Study:

• To develop and validate an explainable, flow-aware deep learning pipeline for joint grading of mitral regurgitation and tricuspid regurgitation.
• To automate view selection, valve assessment, and severity grading from transthoracic echocardiography.
• To enhance the reliability and adoption of AI in routine echocardiographic practice.

Main Methods:

• An end-to-end deep learning pipeline was created, incorporating view selection, leaflet pose estimation, and systolic phase awareness.
• The model was trained on 5086 outpatient transthoracic echocardiography studies and validated on independent cohorts.
• Performance was assessed using AUC, sensitivity, specificity, PPV, and NPV, with subgroup analyses.

Main Results:

• The model achieved excellent discrimination for moderate or greater mitral regurgitation and tricuspid regurgitation (AUC >0.980) with high NPV (>0.970) in internal testing.
• Performance was robust across various demographics and comorbidities, with slight attenuation in specific patient groups.
• External validation showed modestly lower AUCs, particularly for severe regurgitation, but maintained high sensitivity and NPV.

Conclusions:

• An explainable AI framework enables accurate and efficient screening of mitral and tricuspid regurgitation.
• The system demonstrates strong rule-out capability and robust external performance for automated triage.
• This AI tool supports prioritization of echocardiographic studies in diverse clinical settings.