Research on the Identification Method of Respiratory Tract Infections Based on Multi-parameter Fusion

Abstract

Background Cough sounds, as one of the common symptoms of respiratory diseases, are not only rich in information but also translate the health status of the respiratory system. In order to achieve non-invasive, accurate, and rapid identification of respiratory tract infections,this study designed experiments to collect cough sound data from individuals with different symptoms. Methods Various feature parameters were extracted from the cough sound data, and a respiratory tract infection identification model was established based on machine learning methods. A total of 20 patients with upper respiratory tract infections and 37 patients with lower respiratory tract infections were included in this study. Results From these patients, 165 upper respiratory tract infection cough events and 265 lower respiratory tract infection cough events were extracted. MFCC feature parameters, wavelet coefficient-related feature parameters, and entropy-related feature parameters were extracted from the cough events and various combinations were considered. Random forest was employed to construct the respiratory tract infection identification model using different combinations of feature parameters. The results showed that different feature combinations resulted in different modeling performance. Among them, three feature combination strategies showed the best performance, achieving an accuracy of 90% and an AUC of 94%. Conclusions This study successfully realized effective classification of respiratory tract infections through cough sounds, providing doctors with a non-invasive, accurate, and rapid auxiliary diagnostic tool.