Automatic computerized endotracheal tube position verification: an animal model evaluation.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Anesthesiology

Background:

• Improper endotracheal tube placement poses significant risks of morbidity and mortality.
• Accurate verification of endotracheal tube positioning is critical in clinical practice.
• Current methods for verification can be time-consuming and invasive.

Purpose of the Study:

• To develop and evaluate a computer-automated system for endotracheal tube positioning identification using image analysis.
• To assess the accuracy of an algorithm in classifying images of the esophagus, trachea, and carina.
• To create a non-monitoring, processor-based solution for rapid tube placement confirmation.

Main Methods:

• An image analysis algorithm was developed to classify images into esophagus, trachea, and carina categories.
• Image processing involved grayscale conversion and classification based on similarity to pretrained patterns.
• A video sensor mounted on an intubating stylet was used to collect images from 10 bovine throats, with a machine learning approach for algorithm training and testing.

Main Results:

• The algorithm demonstrated high accuracy in classifying anatomical locations: sensitivity of 0.98 and specificity of 0.99 for distinguishing trachea from esophagus.
• Distinguishing between trachea and carina showed a sensitivity of 0.96 and specificity of 0.95.
• Minimal false positives (0.001% esophageal to trachea) and false negatives (0.041% carinal/tracheal to esophagus) were observed.

Conclusions:

• The developed computer-automated system shows high accuracy in verifying endotracheal tube positioning using image analysis.
• The algorithm's performance on nonperfused biological tissue supports its potential as a reliable verification tool.
• Further research is warranted to validate this system for clinical application in endotracheal intubation.