A Scoping Review of Artificial Intelligence in Perioperative Anesthesia: Current Applications, Challenges, and Roadmap for the Future

Author(s): Lisa Le, Ofelia Loani Elvir-Lazo, Robert Wong

Background: Artificial Intelligence (AI) is rapidly integrating into anesthesiology, with the potential to enhance patient safety and advance precision in the field of anesthesia. This scoping review synthesizes recent literature on AI applications across the perioperative care setting.

Methods: A focused review of PubMed, Google Scholar, and Research Rabbit identified 24 articles. The selection included a primary analysis group of 16 articles published in 2025, complemented by 8 earlier studies for context. Findings were thematically synthesized and categorized by perioperative phase: preoperative, intraoperative, and postoperative. AIpowered tools assisted in initial synthesis, followed by manual review and integration by the authors.

Results: The 2025 literature highlights AI’s evolution from theoretical models to clinically relevant tools. Preoperatively, Deep Learning, including Convolutional Neural Networks (CNNs), enables accurate and noninvasive prediction of difficult airways, and comprehensive risk stratification using tools like the POTTER and MySurgeryRisk calculators. Intraoperatively, advanced hybrid models combining Long Short-Term Memory (LSTM), Transformer, and Kolmogorov Arnold Networks (KAN) are being developed for enhanced depth of anesthesia monitoring, while closed-loop systems are used to automate drug delivery. Postoperatively, Natural Language Processing (NLP) and predictive algorithms are utilized to optimize pain management and anticipate complications.

Conclusion: The current literature in advanced AI illustrates a clear progression from theoretical concepts to practical applications in anesthetic care. AI is poised to become an indispensable component of perioperative medicine, facilitating data-driven precision in the field of anesthesia. However, successful and equitable implementation will require addressing key challenges in data governance, model interpretability, and ethical oversight.