Authors: Liu H et al.
BMC Anesthesiology. Published December 11, 2025.
Summary
This study evaluated the development and application of an artificial intelligence–based airway identification system designed to assist tracheal intubation, particularly in settings where operator experience and airway resources are limited. The authors focused on pre-hospital and emergency contexts, where timely airway establishment is critical and often performed by clinicians with variable levels of training.
Using airway images obtained from 978 patients at a single academic center, the investigators created a dataset of 3,912 annotated images through systematic image augmentation. Five deep-learning object detection models were trained to recognize key airway landmarks, specifically the glottic fissure and aryepiglottic folds. Model performance was assessed using standard metrics including precision, recall, F1 score, and mean average precision (mAP).
Among the evaluated models, the YOLO-based architecture demonstrated the strongest performance, with high precision and recall for glottic identification and acceptable performance for aryepiglottic fold recognition. The model achieved a mAP50 of 0.924, indicating robust real-time detection capability. Based on these results, selected models were deployed on a smartphone platform for end-user testing.
In the clinical simulation phase, 72 trainee physicians with no prior intubation experience performed video-laryngoscopic intubation on simulators, either with or without AI assistance. Use of the AI-guided system significantly reduced time to glottic exposure and improved first-attempt success rates. These findings suggest that real-time AI guidance can meaningfully enhance novice performance during airway management tasks.
Overall, the study demonstrates the feasibility of deploying an AI-based airway recognition tool on mobile devices, with potential applications in pre-hospital care, emergency medicine, and training environments. While further validation in real-world clinical settings is needed, this work highlights a promising role for artificial intelligence in augmenting airway management safety and efficiency.
Key Points
An AI-based YOLO model accurately identified key airway anatomical landmarks in real time.
Deployment on mobile devices enabled practical, bedside, and pre-hospital usability.
AI assistance significantly reduced intubation time and improved first-attempt success among novice operators.
This approach may help bridge experience gaps in emergency and resource-limited airway management settings.
Further clinical validation is needed before widespread adoption.
Thank you for allowing us to review and summarize this early-access contribution from BMC Anesthesiology.