Glottic opening detection using deep learning for neonatal intubation with video laryngoscopy

Area of Science:

• Medical technology
• Artificial intelligence in medicine
• Neonatal care

Background:

• Video laryngoscopy (VL) is crucial for neonatal intubation.
• Accurate glottic opening visualization is key for successful intubation.
• Current methods may be limited by user experience and real-time feedback.

Purpose of the Study:

• To develop an AI method to automate glottic opening detection in neonatal VL.
• To enhance the usability and effectiveness of VL for intubation.
• To lay groundwork for future studies on improving intubation outcomes.

Main Methods:

• A deep learning model (YOLOv8) was trained on 1623 video frames from 84 neonatal intubations.
• The model's performance was evaluated using 14-fold cross-validation (precision, recall).
• AI performance was compared against 25 medical providers with varying experience levels.

Main Results:

• The AI achieved 80.8% precision and 75.3% recall in glottic opening detection.
• The AI detected the glottic opening 0.31 seconds faster than the average medical provider.
• AI performance was comparable to novice and intermediate providers, and slightly slower than experts.

Conclusions:

• AI-powered tools can augment VL by offering real-time guidance.
• This technology has the potential to improve neonatal intubation safety and efficiency.
• AI assistance may be particularly beneficial for less experienced medical providers.