Generate Navigations to Guide and Automate Endotracheal Intubation Process

Abstract

In the medical field, Endotracheal Intubation is an emergency procedure that’s performed on people who are unconscious or who cannot breathe on their own. The endotracheal tube which is put through the mouth or nose serves as an open passage through the upper airway. The purpose of Endotracheal Intubation is to permit air to pass freely to and from the lungs to ventilate the lungs. Many complications can occur in the Nasotracheal Intubation process such as bleeding, infections, damages to the voice box (larynx), thyroid gland, vocal cords and windpipe (trachea), or pharynx. To overcome the consequences, we propose a system to guide and automate endotracheal intubation by providing the physician with necessary navigations for moving the intubation tube. This study is a proof of concept to prove that generating navigations using a trained model based on Convolution Neural Network has practical potential. In the proposed system, different anatomical landmarks along the path from nose to carina are identified, captured images are preprocessed and use a Convolution Neural Network based on image classification to predict the class of each anatomical landmark. The proposed system has been trained using the training dataset of 9753 images, and the performance of the system was evaluated using the testing dataset of 2438 images. The results, obtained using the pre-trained model, show that the system correctly classified 2385 out of 2438 (97.8%).