Artificial Intelligence in Echocardiography: A Critical Review of Models, Validation, and Clinical Impact.

OBJECTIVE: To critically review artificial intelligence (AI) in echocardiography, linking model design and technical performance to validation and clinical use. METHODS: We synthesized literature assembled through Web of Science and Scopus collections and targeted revision searches. Evidence was organized by echocardiographic task, model paradigm, data source, evaluation protocol, and level of clinical validation. RESULTS: AI applications span acquisition guidance, image analysis, quantitative measurement, disease assessment, and outcome prediction, with emerging foundation, multimodal, and examination-level models. Most studies assess retrospective task performance; fewer evaluate cross-site transportability or clinical impact. Reproducible performance has been demonstrated for view recognition, segmentation, and selected measurements under aligned protocols, and prospective trials confirm workflow gains. However, evidence linking AI to improved clinical decisions or patient outcomes remains limited. Generalizability is further constrained by uneven dataset coverage and inconsistent evaluation protocols. CONCLUSION: Current evidence supports AI assistance in selected echocardiographic tasks; however, clinical use requires external validation, failure reporting, and ongoing monitoring in intended settings. SIGNIFICANCE: This review links model development with validation evidence to guide clinical evaluation and identify priorities for prospective studies in echocardiography.