Artificial intelligence in cardiopulmonary resuscitation: a systematic review

Cardiac arrest remains a major cause of mortality and neurological disability, and its management depends on rapid recognition, effective resuscitation, and accurate post-arrest prognostication. Artificial intelligence (AI) has been increasingly investigated as a tool to support these stages of care. This narrative review summarizes recent evidence on AI applications in cardiac arrest, including prediction, recognition, intra-arrest support, prognostication, and cardiovascular risk stratification. Searches were conducted in PubMed/MEDLINE, Google Scholar, EMBASE, Scopus, IEEE Xplore, and ScienceDirect for studies published from January 2000 to June 2025. Studies addressing AI, machine learning, or deep learning in cardiac arrest-related contexts were descriptively synthesized according to clinical application, AI methodology, and care setting. Overall, 59 studies were included. Six application categories were identified: early prediction of cardiac arrest or clinical deterioration (n = 16), cardiac arrest recognition and emergency response activation (n = 3), intra-arrest CPR and defibrillation support (n = 6), post-cardiac arrest prognostication (n = 20), risk stratification for sudden cardiac death and malignant arrhythmias (n = 11), and related cardiovascular or perioperative prognostic models (n = 3). Deep learning and neural-network-based approaches were the most frequent methodologies (n = 35), followed by traditional machine learning (n = 18) and natural language processing (n = 6). AI showed particular promise for ECG-based prediction, temporal clinical trajectories, emergency call interpretation, waveform and defibrillation analyses, and neurological or mortality prognostication after cardiac arrest. However, the evidence remains heterogeneous, with many studies relying on retrospective datasets, selected populations, and limited external or prospective validation.