Automated Assessment of Difficult Airway With Facial Recognition Techniques

Overview

General anaesthesia mandates artificial ventilation and tracheal intubation in order to provide patients with artificial breathing. Difficulties related to ventilation and intubation remain the leading cause of morbidity and mortality in general anaesthesia, essentially due to inaccuracies in pre-operative detection of anatomical factors predisposing to difficult airways. In this project investigators will develop image and video-processing technologies software solutions to allow automatic recognition of anatomical features playing a key role in identification of difficult ventilation and intubation, leading to modifications in pre-operative anaesthesia management assessment and therefore increase patients' safety.

Any tracheal intubation requires a pre-operative screening and assessment in order to obtain the essential medical history of the patient, optimize patients' condition in case of any co-existing disease before the operation and select the best method of anesthesia for the day of surgery. The aim of this assessment is to identify potential anesthetic difficulties, such as predictors of difficult airways, which still nowadays represent the first cause of litigation in anesthesia related closed claim studies. In the first step of the pre-operative assessment procedure, the patient will be analyzed by the software. The patient will be automatically guided through a 10 minutes series of tests and the software will analyze in real-time his/her morphological and dynamic features in order to classify the patient into one of 5 categories described in the next Section. Details relevant to difficult ventilation and intubation (static and dynamic), such as quantifying the exact inter-incisors distance (mouth opening), visibility and detection of anatomical landmarks in the open mouth (uvulae, pillars, tonsils, tongue, posterior pharynx), thyro-mental distance, neck circumference, neck mobility with maximal anterior and posterior movement. The analysis will be performed by: * automatically computing these relevant measures using robust computer vision algorithms capable to detect, describe and track the face and the neck with high level of accuracy and robustness to extreme poses (left and right rotation and up and down movement of the face) * developing powerful image processing techniques to describe and compute intra-oral structures. The two sets of measures will be then combined into a machine learning approach capable to classify the patient. The results of the analysis as well as all the recorded videos of every single test will be stored on a central database and accessed in real-time by the doctor to continue the pre-operative consultancy. The patient will then undergo his planned surgery at the initially planned time and be intubated for that purpose. Proper recording of the grade of intubation in the operating room will be documented and introduced in the assessment database. By this mean, the database will evolve with the assessment and the final post-operative intubation score so that to improve the automatic predictability of the machine learning algorithm.