Deep-learning model associating lateral cervical radiographic features with Cormack-Lehane grade 3 or 4 glottic view.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Anesthesiology

Background:

• Unanticipated difficult laryngoscopy poses significant risks for airway-related complications.
• Predicting difficult laryngoscopy is crucial for patient safety during surgical procedures.

Purpose of the Study:

• To develop and validate a deep-learning model utilizing lateral cervical spine radiographs.
• To predict Cormack-Lehane grade 3 or 4 direct laryngoscopy views of the glottis.

Main Methods:

• A convolutional neural network (CNN)-based deep-learning model was developed.
• The model was trained and tested on 5939 lateral cervical spine radiographs from thyroid surgery patients.
• Performance was compared against six established deep-learning models (VGG, ResNet, Xception, ResNext, DenseNet, SENet).

Main Results:

• The developed model achieved a superior Brier score (0.023) compared to all other models (p < 0.001).
• Key performance metrics included R² (0.428), mean squared error (0.023), mean absolute error (0.048), balanced accuracy (0.713), and AUC (0.965).
• Radiographic features of the hyoid bone, pharynx, and cervical spine correlated with difficult glottic views.

Conclusions:

• The novel deep-learning model demonstrates high accuracy in predicting difficult laryngoscopy views from cervical spine radiographs.
• This AI-driven approach offers a promising tool for pre-operative risk assessment and improved patient safety.
• Further research can explore integration into clinical workflows for real-time airway management support.