Predicting Difficult Tracheal Intubation Using Multi-Angle Photographic Analysis with Convolutional Neural Networks...

Area of Science:

• Anesthesiology
• Artificial Intelligence
• Medical Imaging

Background:

• Difficult tracheal intubation poses significant risks for anesthesia-related morbidity.
• The incidence of difficult airways ranges from 1% to 4% in clinical practice.
• Current methods for predicting difficult airways have limitations in sensitivity and standardization.

Purpose of the Study:

• To investigate the potential of artificial intelligence-driven photographic analysis for predicting difficult tracheal intubation.
• To develop a novel, accessible tool for preoperative airway assessment.

Main Methods:

• Preoperative photographs were taken from patients in various positions and with different facial configurations.
• A convolutional neural network (CNN) model was trained to analyze these images.
• Intubation difficulty was categorized based on the number of attempts and interventions required.

Main Results:

• The CNN model demonstrated effective performance in predicting difficult intubation.
• Optimal performance was achieved with specific batch sizes (32, 128) and learning rates (0.1 to 0.001).

Conclusions:

• AI-based photographic analysis using a mobile phone can successfully predict difficult tracheal intubation at the bedside.
• This approach offers a promising, easily accessible alternative to current prediction methods.
• The study provides a new perspective for estimating intubation difficulty in anesthesia.