Table 5 AI Models in Mitral Regurgitation (MR) Evaluation
From: Contemporary applications of artificial intelligence and machine learning in echocardiography
Study | Dataset/Population | Algorithm/AI Method | Application | Key Results | AUROC/Accuracy |
|---|---|---|---|---|---|
Hausleiter et al. (EuroSMR) | 4600 patients (4172 derivation, 428 validation) | ML risk score with SHAP explainability | Mortality risk prediction for M-TEER | Identified high-risk patients with >70% mortality | AUC 0.789 |
Sadeghpour et al. | Multicenter observational cohorts | Multiparametric ML model | MR severity grading | 80% accuracy for severity; 97% for moderate/severe MR | Accuracy 80–97% |
Bernard et al. | 400 patients (France + Canada) | Explainable AI + hierarchical clustering | Phenogrouping for MVS benefit prediction | C-index 0.75; improved reclassification (NRI P = 0.002) | — |
Yang et al. | 2080 studies from multiple hospitals | CNN + sequence model | Multivalvular disease detection (incl. MR) | MR AUC 0.88; comparable to expert readers | AUC 0.88 |
Edwards et al. | Pediatric echocardiograms, PLAX views | Two CNNs (view + MR detection) | Pediatric MR detection | High accuracy despite image variability | AUROC 0.91 |
Martins et al. | 11,646 videos from 912 exams | 3D CNN + decision tree meta-classifier | RHD diagnosis including MR | Best performance in Definite RHD (85.8%) | Accuracy 72.77% |
Peck et al. | 36 novices, 50 patients | AI-guided imaging tool | MR image acquisition by novices | 90% diagnostic-quality acquisition | — |
Brown et al. | 511 pediatric echoes | SVM + Transformer + 3D CNN | MR jet length analysis for RHD | Close match with expert grading | AUC 0.93 |
Trenkwalder et al. | Multicenter MR registry | Unsupervised ML clustering | MR phenotyping for TEER outcome stratification | Identified subgroups with distinct remodeling/fibrosis | — |
