Improving difficult direct laryngoscopy prediction using deep learning and minimal image analysis: a single-center...

Area of Science:

• Anesthesiology
• Artificial Intelligence
• Medical Imaging

Background:

• Predicting difficult direct laryngoscopy (DDL) is crucial for patient safety during airway management.
• Current methods for predicting DDL may be invasive or time-consuming.
• A non-invasive, easily accessible method for DDL prediction is needed.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for predicting DDL.
• To utilize smartphone-captured facial and neck images for DDL prediction.
• To assess the model's performance and identify key predictive features.

Main Methods:

• A prospective, single-center study included adult patients undergoing endotracheal intubation.
• A deep learning model (EfficientNet-B5) was trained on smartphone images (frontal, lateral, neck extension, open mouth views).
• Multitask learning incorporated picture view information; model performance was evaluated using AUC and F1-score.

Main Results:

• The AI model achieved a receiver operating characteristic area under the curve (AUC) of 0.81-0.88.
• The model's F1-score ranged from 0.72-0.81 for difficult direct laryngoscopy prediction.
• Analysis indicated that neck and chin features in frontal and lateral views significantly contribute to prediction accuracy.

Conclusions:

• A deep learning model using smartphone images can effectively predict difficult direct laryngoscopy.
• The AI model offers a practical, non-invasive, and easily implementable approach to airway management assessment.
• Incorporating multiple image views enhances the predictive performance of the AI model.