Artificial intelligence for early detection and prevention of ventilator-induced lung injury: a scoping review

Abstract Background The integration of artificial intelligence (AI) into intensive care unit (ICU) ventilator management has attracted growing research interest, with studies exploring applications ranging from surveillance and decision support toward more autonomous ventilatory adjustment. Whether this represents a clinically validated transition from reactive to proactive ventilation management remains to be established through prospective evidence. This review maps the available literature on these developments in the context of ventilator-induced lung injury (VILI), a heterogeneous syndrome including barotrauma, volutrauma, atelectrauma, and biotrauma. This review maps AI applications relevant to VILI detection and prevention, rather than studies that directly demonstrate VILI prevention. Methods This scoping review mapped and synthesized evidence on the use of AI-based systems for the early detection and prevention of VILI in critically ill adults receiving invasive mechanical ventilation. Following Joanna Briggs Institute methodology and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines, a systematic search was conducted across PubMed, Scopus, Web of Science, ScienceDirect, Google Scholar, and the Biblioteca Virtual en Salud (BVS) for literature published between 2010 and 2026. Results A total of 52 studies met full-text inclusion criteria. Most AI applications focused on automated detection of patient-ventilator asynchronies, prediction of complications related to acute respiratory distress syndrome (ARDS), and optimization of lung-protective ventilation parameters through machine learning and reinforcement learning approaches. Although AI-based systems may support more individualized ventilatory management strategies, their clinical implementation remains limited by the lack of external validation, methodological heterogeneity, and scarce evidence in low- and middle-income settings. Conclusions The evidence identified is predominantly technical, retrospective and exploratory, and rests largely on surrogate physiological or classification endpoints. High diagnostic accuracy, or improvement in surrogate parameters, does not by itself constitute evidence that these systems prevent VILI or improve patient-centred outcomes. Current evidence therefore supports the promise of these applications and the prioritization of prospective research, but not their routine clinical implementation.