Identification of Elderly Patients with Lower Respiratory Tract Infection by Artificial Intelligence Analysis of Cough Pattern Sounds.

Abstract

Automatic recognition of cough sounds shows promise in the diagnosis of respiratory conditions. This study investigated the diagnostic value of cough sounds in elderly patients with lower respiratory tract infection (LRTI). We selected 83 elderly patients with suspected LRTI who sought medical advice at our hospital from January 2022 to September 2022, and grouped them into the infected and uninfected categories, according to their clinical traits. The cough sound of each subject was recorded and features were extracted using the Mel Frequency Cepstrum Coefficient. Four cough sound indexes, including the length of light or heavy cough time (T1), frequency of sound, decibels full scale (dBFs) and total length of cough time (T0) were compared between the two groups. The diagnostic efficacy of each index was analyzed using the receiver operating characteristic (ROC) curve. 22 patients were diagnosed with LRTI in the infected group including 15 males and 7 females, 13 were in the LRTI-free uninfected group, including 7 males and 6 females. Cough sound indexes were higher in the infected group compared with the uninfected group at T1 (p = 0.127), frequency of sound (p = 0.894), dBFs (p = 0.532) and T0 (p = 0.854). ROC curve analysis showed that the area under the curve (AUC) values of the above four indexes and the combined indexes for LRTI diagnosis were 0.680, 0.503, 0.577, 0.486 and 0.696, respectively. Cough sounds are correlated with LRTI. However, due to the small sample size of this study, the current results do not find that automatic recognition of cough has obvious diagnostic value, but its diagnostic potential in elderly patients with LRTI cannot be denied.