Artificial Intelligence and Exercise-Based Rehabilitation in Adults with Burn Injury: A Scoping Review of Emerging Evidence and Implications for Nursing Practice

Abstract

Background: Post-burn rehabilitation requires resistance and isometric training to reduce muscle wasting and contractures. Artificial Intelligence (AI) and Machine Learning (ML) are increasingly used to monitor and optimize rehabilitation. However, their clinical, legal, and long-term implications remain poorly understood.
Objective: To synthesize evidence on AI/ML applications in resistance and isometric training for burn rehabilitation, focusing on clinical efficacy, operational approaches, and research gaps.
Methods: A scoping search of PubMed, Scopus, and Web of Science identified peer-reviewed studies published from 2020–2026 using terms related to burns, AI, resistance exercise, and rehabilitation. Following PRISMA 2020 guidelines, 15 studies were included in the qualitative synthesis.
Results: Studies used Convolutional Neural Networks for pose estimation, predictive models for phenotypic monitoring, and Long Short-Term Memory networks for surface electromyography fatigue monitoring. Across the included studies that reported adherence, adherence estimates varied substantially according to the intervention, monitoring method, and follow-up period. Several studies reported improved exercise engagement or adherence with technology-assisted monitoring; however, the heterogeneity of measurement approaches and study designs precluded calculation of a reliable pooled effect estimate.
Conclusion: AI-enabled technologies show emerging potential for movement assessment, exercise monitoring, fatigue detection, and individualized rehabilitation support in adult burn rehabilitation. However, the evidence remains limited and heterogeneous, with few high-quality clinical studies directly evaluating AI-integrated exercise interventions. Future research should prioritize multicenter prospective studies, standardized outcome measures, transparent reporting of model performance, and human-in-the-loop safety and governance frameworks.