Design and Development of an On-Line Acoustic-Guided Endotracheal Intubation Assistive System for Glottic Opening Detection

  • 陳 威豪

Student thesis: Doctoral Thesis


Accomplishing successful endotracheal intubation (ETI) both rapidly and with a minimum number of attempts is a crucial skill for clinicians in order to manage patient airways and acute care Repeated attempts may cause airway injury and increase the occurrence of hypoxemia regurgitation or aspiration of gastric contents hemodynamic instability and even death Successful ETI within minimum attempts is highly dependent on clinician experience Most current devices help confirm endotracheal tube position after intubation but are not useful in reducing unnecessary attempts It would be highly beneficial to develop an objective glottis identification method during ETI to minimize unnecessary intubation attempts The purpose of this research was to develop an innovative acoustic-guided tracheal intubation assistive system to help ETI providers judge among the glottis and other hypopharyngeal structures during the procedure By insufflating oxygen into hypopharynx the acoustic responses of non-glottic structures and the glottis were analyzed by the proposed algorithm to recognize the glottis and determine the glottic boundary More specifically the research describes 1) development of a PC-based acoustic-guided tracheal intubation assistive system including oxygen insufflation device electronic stethoscope system and laptop with acoustic signal analysis software; 2) system calibration and analysis in response to oxygen insufflation of hypopharyngeal structures; 3) development of an on-line acoustic-guided glottis identification algorithm to investigate the feasibility in situ using live ETI cases In the first study the feasibility of acoustic analysis for glottis discrimination was investigated by using linear prediction coefficients (LPCs) and mel frequency ceptstral coefficients (MFCCs) as the representative sound features in linear discriminant analysis (LDA) and Gaussian mixture model (GMM) methods In the second study to simulate the glottis searching process during ETI a GMM-based likelihood ratio algorithm was developed to determine the acoustic evaluated boundary between non-glottic and glottic segments in the sound recordings In the third study in order to accomplish on-line glottis recognition a two-stage segment-based acoustic approach including GMM-based delta Bayesian information criterion (delta-BIC) and majority vote was applied for fast glottis identification The results of the first study showed the proposed GMM-based classifier outperformed the conventional LDA method in accuracy (92 34% vs 86 35%) at the mixture number of 3 The results of the second study showed the actual boundaries in all 9 cases were successfully detected by the GMM-based likelihood ratio method The time differences between the evaluated boundary and actual boundary were less than 416 msec The result of the third study using the proposed two-stage segmented-based acoustic approach showed the success rate of glottic boundary recognition was 77 78% for a single screening attempt The time differences between the evaluated and actual boundary were less than 384 msec This series of studies demonstrate the acoustic-guided tracheal intubation assistive system developed in this work is likely to be feasible for on-line glottis and glottic boundary detection during ETI The eventual goal is to develop an innovative objective monitor to increase initial endotracheal intubation success rate minimize unnecessary intubation attempts and increase patient safety for airway management
Date of Award2015 Jan 27
Original languageEnglish
SupervisorKuo-Sheng Cheng (Supervisor)

Cite this