Background: Difficult Airway (DA) is a condition where anesthesiologists face challenges in mask ventilation or intubation, leading to serious injuries and a significant portion of anesthesia-related mortality. Traditional assessments involve medical history and clinical examinations to identify potential difficulties. Recently, machine learning has become a valuable tool in predicting difficult intubation by analyzing various morphological and clinical variables. This review aims to explore the machine learning approaches used, the predictors identified, and the accuracy of different models in this area. Materials and Methods: This study utilized keywords like Machine Learning, Transfer Learning, and Intratra-cheal Intubation, searching databases such as PubMed, Cochrane, and Google Scholar from ۲۰۰۰ to November ۲۰۲۴. Exclusions included studies without full-text access, ambiguous results, and non-reputable sources. Results: Initially, ۹۱ articles were identified and organized in Endnote X۱۶, which was reduced to ۵۱ after removing duplicates. Following an evaluation of titles and abstracts, ۳۲ articles were excluded. Authors of articles lacking full-text access were contacted, but ultimately, only ۹ articles were selected for the study. The systematic review of these ۹ articles revealed that various machine learning approaches are effective in predicting difficult laryngoscopy and intubation. Models such as KNN, XGBoost, random forests, SVM, and the J۴۸ decision tree showed high accuracy by analyzing demographic features. Key predictors included morphological features like patient images, voices, and clinical variables. The accuracy of machine learning models in predicting difficult intubation ranged from ۷۰% to ۹۵%, varying based on data quality and feature selection. Decision tree-based models excelled in identifying at-risk patients, while gradient-boosting models like XGBoost were better at recognizing complex patterns. Conclusion: In summary, the use of machine learning in predicting difficult laryngoscopy and intubation can contribute to improved clinical outcomes. However, futher research is needed in this area to optimize models and enhance prediction accuracy. Additionally, examining the impact of various variables on model performance can aid in the development of better clinical tools.
