Tracheal Opening Discrimination During Intubation Using Acoustic Features and Gaussian Mixture Model

10.5405/jmbe.1714

Wei-Hao Chen；Yu-Hsien Chiu；Hsien-Chang Wang；Yu-Wei Hung；Hao-Po Su；Kuo-Sheng Cheng

Airway management ； Endotracheal intubation ； Gaussian mixture model ； Mel-frequency cepstral coefficients

Journal of Medical and Biological Engineering

34卷6期（2014 / 12 / 01）

605 - 611

英文

Correct identification of glottic opening is crucial during endotracheal tube intubation for airway management. Direct glottis visualization by the physician is considered best standard practice, but it is dependent on conditions, skill, and experience. This study proposes an improved oxygen insufflation system that applies an acoustic modeling approach to discriminate glottic and non-glottic areas utilizing the acoustic response from a steady directional airflow at the hypopharynx. An electric stethoscope is placed at the suprasternal notch to record sound produced by insufflation during intubation. The Gaussian mixture model with mel-frequency cepstral coefficients (MFCCs) is used to determine differences between glottic and non-glottic areas. A dataset containing 56.567 seconds of non-glottic sound and 46.472 seconds of glottic sound was recorded from 8 anesthetized adults receiving intubation. Short-time analysis and several objective evaluations were performed to investigate system performance. The evaluation results show that the system achieved a high classification accuracy of 93.24%. The proposed approach outperformed a baseline linear discriminant analysis method with various configurations of linear prediction coding coefficients and MFCCs, and shows potential in improving glottic identification during endotracheal tube intubation.