A Bayesian Classifier Combination Methodology for Early Detection of Endotracheal Obstruction of COVID-19 Patients in ICU

ABSTRACT

Due to COVID-19 related complications, many of the diagnosed patients end up needing intensive care. Complications are often severe, to such an extent that mortality rates in these patients may be high. Among the wide variety of complications, we find necrotizing tracheobronchitis, which appears suddenly with the obstruction of the endotracheal tube. This complication can cause severe damage to the patient or even death. In order to help clinicians with the management of this situation, we propose a Machine Learning-based methodology for detecting and anticipating the obstruction phenomenon. Through the use of Bayesian classifiers, classifier combination, morphological filtering and a track-while-scan detection mode we are able to establish an indicator function that serves as a reference to clinicians. Our experiments show promising results and lay the foundations of an intelligent system for early detection of endotracheal obstruction. © 2021, Springer Nature Switzerland AG.