A Deep Learning Method to Predict Difficult Laryngoscopy Using Cervical Spine X-ray Image

Predicting a difficulty of a laryngoscopy is important for patient safety, as an unanticipated difficult laryngoscopy is associated with serious airway-related complications, such as brain damage, cardiopulmonary arrest, or death. Although clinical predictors, such as the modified Mallampati classification, thyromental distance, inter-incisor gap, and the upper lip bite test, are used for airway evaluation in clinical practice, these indicators have low sensitivity and large inter-assessor variability and require patient cooperation. The investigators developed a deep learning-based model that predicts a difficult laryngoscopy from a cervical spine lateral X-ray using data from 14,135 patients undergoing thyroid surgery. And this study is under submission. This deep learning model showed the highest performance in predicting difficult laryngoscopy compared to other deep learning models (VGG-Net, ResNet, Xception, ResNext, DenseNet, and SENet) with a sensitivity of 95.6%, a specificity of 91.2%, and an area under ROC curve (AUROC) of 0.972. However, as the model was a retrospective design using existing medical records, the presence or absence of cricoid pressure to obtain the optimal laryngoscopy was not evaluated, and not compared with airway evaluations. In this study, the investigators prospectively validate the model for predicting a difficult laryngoscopy and compare predictive power with clinical airway evaluation. If this study prospective confirm our results, this approach can be helpful in improving patient safety and preventing airway-related complications through objective and accurate airway evaluation.