VARAL: Vision-Based Autonomous Recognition of Airway Landmarks for Endotracheal Intubation

Autonomous recognition of anatomical landmarks in the airway plays a significant role in assisting clinicians to perform a video-guided endotracheal intubation (ETI) rapidly and accurately, which can increase its efficiency and success rate. However, existing recognition models still can not identify all anatomical landmarks of the airway, and their clinical assistance effects have not been quantitatively evaluated, which seriously restricts their clinical applications. This study proposes a vision-based autonomous recognition system for airway landmarks (VARAL system) based on YOLOv5 structure to assist in video-guided ETI, which can identify all the landmarks in the airway, including uvula, epiglottis, arytenoid cartilage and glottis. On the real clinical ETI dataset, this model performs the best in recognizing epiglottis and glottis compared with previous studies, which have received widespread attention. It also performances well in recognizing uvula and arytenoid cartilage, which have been rarely studied. Assessment of clinical assistance effect is conducted, which involves ten junior doctors (have completed fewer than 50 independent intubation cases) and eleven senior doctors (have completed more than 500 independent intubation cases). Results illustrate that the overall performance of the model outperforms junior doctors and is comparable to senior doctors. This proposed model can assist clinicians in obtaining more comprehensive information of the ETI pathway, enhancing the safety of the ETI and strengthening the navigation of ETI robots. The results of the clinical assistance assessment provide preliminary evidence for future clinical applications.