Artificial intelligence models, particularly those trained on echocardiographic imaging, demonstrated strong discriminatory performance for mitral regurgitation evaluation with reported AUCs ranging from 0.74 to 0.94.
Systematic Review (n=80,915)
Do artificial intelligence models improve the diagnosis, severity assessment, and prognostication of mitral regurgitation in patients with or at risk for the disease?
Artificial intelligence models, particularly those trained on echocardiographic data, demonstrate strong potential for enhancing the diagnostic and prognostic evaluation of mitral regurgitation, though external validation is required for clinical adoption.
Effect estimate: AUC 0.74-0.94
BACKGROUND: Mitral regurgitation (MR), one of the most common valvular heart diseases, poses ongoing challenges in risk stratification and timely intervention. Traditional diagnostic approaches, suffer from interobserver variability. Artificial intelligence (AI) has recently gained traction in cardiology to augment clinical precision. This article reviews the use of AI in the diagnosis, severity assessment, and prognostication of MR, with a focus on performance metrics. METHODS: The search was conducted in PubMed, Embase, and MEDLINE on May 9, 2025. Studies were eligible if they applied AI to MR-related tasks using imaging, ECG, or clinical data. Data extraction focused on dataset characteristics, model architectures, and performance. RESULTS: A total of eleven studies, comprising 80,915 patients, were included. Among the included studies, six utilized echocardiographic data, two electrocardiography, two clinical biomarkers or structured datasets, and one chest radiography. Algorithms included convolutional neural networks, support vector machines, and ensemble models. Reported AUCs ranged from 0.74 to 0.94. Models based on color Doppler or 3D geometrical mitral features achieved the highest discriminatory performance. Only a minority of studies incorporated external validation or reported clinically actionable thresholds such as PPV. ECG-based models demonstrated high scalability but lower sensitivity. Studies integrating multimodal data yielded promising results. CONCLUSION: AI models, especially those trained on echocardiographic imaging, demonstrate strong potential for improving MR evaluation. However, widespread clinical adoption is limited by lack of external validation, and inconsistent outcome reporting. Future work should emphasize model interpretability, multicenter validation, and head-to-head comparisons with expert assessment to bridge the translational gap.
Ahmadi-Hadad et al. (Thu,) conducted a systematic review in Mitral regurgitation (n=80,915). Artificial intelligence models was evaluated on AI performance score (diagnostic accuracy and prognostic capability) (AUC 0.74-0.94). Artificial intelligence models, particularly those trained on echocardiographic imaging, demonstrated strong discriminatory performance for mitral regurgitation evaluation with reported AUCs ranging from 0.74 to 0.94.