Ricci Fabrizio, Rizzuto Maria Luana, Bisaccia Giandomenico, Mansour Davide, Gallina Sabina, Sciarra Luigi, Bagliani Giuseppe, Dello Russo Antonio, Mortara Andrea, Ciliberti Giuseppe
Dipartimento di Neuroscienze, Imaging e Scienze Cliniche, Università degli Studi "G. d'Annunzio" di Chieti-Pescara - U.O.S.D. Cardiologia Universitaria, Dipartimento Cuore, ASL 2 Regione Abruzzo, Chieti.
Dipartimento di Neuroscienze, Imaging e Scienze Cliniche, Università degli Studi "G. d'Annunzio" di Chieti-Pescara.
G Ital Cardiol (Rome). 2025 Sep;26(9):635-646. doi: 10.1714/4542.45427.
Artificial intelligence (AI) is redefining ECG interpretation, transforming it from a static diagnostic tool into a dynamic, predictive, and integrative instrument. Although widespread, traditional rule-based ECG analysis has limitations in accuracy and adaptability, especially in complex clinical settings. In contrast, AI-driven models, particularly those employing machine learning and deep learning architectures, have demonstrated improved diagnostic performance across a broad spectrum of cardiovascular diseases, including atrial fibrillation, acute myocardial infarction, hypertrophic cardiomyopathy, and valvular heart disease. Notably, AI-ECG is now able to detect subclinical ventricular dysfunction, stratify long-term risk, and anticipate major adverse events before overt clinical manifestations occur. In addition to diagnosis, AI-ECG is emerging as a decision support tool in scenarios characterized by diagnostic uncertainty, such as syncope and cardio-oncology, and may significantly optimize triage and resource allocation. Multiparametric approaches further extend its utility, enabling simultaneous prediction of structural, functional, and electrical cardiac parameters. Wearable devices integrated with AI improve continuous monitoring and may decentralize arrhythmia detection and sudden cardiac death prevention. Despite these advances, critical challenges remain. Poorly explainable AI models, algorithmic bias, overfitting, data governance, and regulatory uncertainty demand rigorous methodological scrutiny. In this framework, federated learning architectures may enable continuous multicenter model refinement and enhance methodological robustness while safeguarding data privacy. The European AI Act and methodological checklists promoted by scientific societies offer a framework to address these issues, fostering transparency, equity, and clinical validity. If validated and implemented responsibly, AI-enhanced ECG has the potential to enhance - not replace - clinical reasoning, advancing a precision medicine paradigm based on both technological innovation and human expertise.
人工智能（AI）正在重新定义心电图解读，将其从一种静态诊断工具转变为动态、预测性和综合性工具。尽管传统的基于规则的心电图分析广泛应用，但在准确性和适应性方面存在局限性，尤其是在复杂的临床环境中。相比之下，人工智能驱动的模型，特别是那些采用机器学习和深度学习架构的模型，在包括心房颤动、急性心肌梗死、肥厚型心肌病和心脏瓣膜病在内的广泛心血管疾病中，已显示出改善的诊断性能。值得注意的是，人工智能心电图现在能够检测亚临床心室功能障碍，分层长期风险，并在明显临床表现出现之前预测重大不良事件。除了诊断之外，人工智能心电图正在成为诊断不确定性场景（如晕厥和心脏肿瘤学）中的决策支持工具，并可能显著优化分诊和资源分配。多参数方法进一步扩展了其效用，能够同时预测心脏的结构、功能和电参数。与人工智能集成的可穿戴设备改善了连续监测，并可能分散心律失常检测和心脏性猝死预防。尽管取得了这些进展，但关键挑战仍然存在。难以解释的人工智能模型、算法偏差、过拟合、数据治理和监管不确定性需要严格的方法学审查。在此框架下，联邦学习架构可以在保障数据隐私的同时，实现多中心模型的持续优化，并提高方法的稳健性。欧洲人工智能法案以及科学协会推广的方法学清单提供了一个解决这些问题的框架，促进透明度、公平性和临床有效性。如果经过验证并负责任地实施，人工智能增强的心电图有潜力增强而非取代临床推理，推动基于技术创新和人类专业知识的精准医学范式发展。
