Artificial Intelligence System for the Prediction of Difficult Airway

Abstract

A precise evaluation of difficult airway (DA) is crucial in anesthesiology. Artificial intelligence (AI) emerges as a promising tool, but its effectiveness in predicting DA is still unclear. Objective: To demonstrate the effectiveness of the AI system for predicting difficult airway. Method: A comparative, prospective, cross-sectional observational study. Fifty patients were evaluated using standardized photographs analyzed by AI and conventional airway (AW) assessment by anesthesiologists. Results: The AI method identified a DA incidence of 16%, while the conventional method detected 38%. AI showed a sensitivity of 42%, specificity of 94%, positive predictive value of 80%, and negative predictive value of 73%. The area under the ROC curve was 0.6782683 for AI and 1 for the conventional method. While AI demonstrated high specificity, its moderate sensitivity suggests limitations in predicting DA. The conventional method showed superior performance. Conclusion: AI has potential as a complementary tool in predicting DA, but it currently cannot replace conventional clinical assessment.