A Novel Artificial Intelligence System for Endotracheal Intubation

Area of Science:

• Medical technology
• Artificial intelligence in medicine
• Emergency medicine

Background:

• Successful endotracheal intubation (ETI) relies on visualizing the glottic opening.
• Objective tools for real-time guidance during ETI are limited.
• The utility of artificial intelligence (AI) for ETI guidance is unexplored.

Purpose of the Study:

• To evaluate the accuracy of various AI algorithms in identifying the glottic opening.
• To develop a tool that could assist in successful ETI.
• To assess the potential of AI in real-time ETI guidance.

Main Methods:

• Seven providers performed 10 ETI attempts each on a mannequin using video laryngoscopy.
• AI algorithms (KNN, SVM, decision trees, NN) analyzed recorded ETI attempts.
• Algorithms were trained on half the videos and tested on the remaining half for accuracy, sensitivity, and specificity.

Main Results:

• AI algorithms demonstrated varying performance in glottic opening detection.
• K-nearest neighbor (KNN) and support vector machine (SVM) achieved 70% sensitivity and 90% specificity.
• Neural networks (NN) showed 72% sensitivity and 78% specificity.

Conclusions:

• Initial AI algorithms can identify the glottic opening with over 80% accuracy.
• Further AI development could enable real-time feedback for ETI.
• AI holds potential to improve ETI success rates through guided assistance.