The respiratory sound features of COVID-19 patients fill gaps between clinical data and screening methods

Ying hui Huang; Si jun Meng; Yi Zhang; Shui sheng Wu; Yu Zhang; Ya wei Zhang; Yi xiang Ye; Qi feng Wei; Nian gui Zhao; Jian ping Jiang; Xiao ying Ji; Chun xia Zhou; Chao Zheng; Wen Zhang; Li zhong Xie; Yong chao Hu; Jian quan He; Jian Chen; Wang yue Wang; Chang hua Zhang; Liming Cao; Wen Xu; Yunhong Lei; Zheng hua Jian; Wei ping Hu; Wen juan Qin; Wan yu Wang; Yu long He; Hang Xiao; Xiao fang Zheng; Yi Qun Hu; Wen Sheng Pan; Jian feng Cai

The audio analysis was conducted by six respiratory specialists having 5 to 23 years of experience. The six doctors are from five hospitals, including NPFH, the Seventh Affiliated Hospital of Sun Yatsen University, Pucheng County Hospital of Traditional Chinese Medicine, First Affiliated Hospital of XiaMen University, and the Second Afficiated Hospital of Xiamen Medical College. There was no communication between these physicians before and after auscultation. The content of auscultation assessment includes breathing rate, intensity, and whether there were abnormal or additional breath sounds. Each audio was played over and over again until the physician made a decision. All the results of each physician were recorded in sheet, then summarized and normalized by two graduate students who were blinded to patient information. The terminology proposed by the ad hoc committee of the International Lung Sounds Association was used to agree on the respiratory sounds described by six physicians. In addition to the result judgments of respiratory doctors, the above audio is also graphically displayed and quantitatively described by means of signal analysis using Matlab R2019, v9.6.0 (MathWorks, USA). The relationship between signal characteristics and time is intuitively demonstrated by drawing waveform and time-frequency diagram. By observing the waveform of breath sound, the duration of inhalation and breath as well as the intensity of sound can be preliminarily understood. The main frequency range of normal and abnormal breath sounds can be visualized by detecting the time-frequency chart. The background noise generated by the movement between the patient and the stethoscope is a high energy artifact on the time-frequency graph, which can be easily recognized in the image. The feature of normal respiratory sounds and pneumonic respiratory sounds refer to APSP project 10 . The images were excluded from the analysis. Then the corresponding spectrum diagram was drawn by fast Fourier transform (FFT), which can analyze the signal strength of a particular frequency.