Automatic Analysis of Lung Sounds in 3-Year-Old Children.

Abstract

Introduction: In recent years, with the advent of artificial intelligence, clear progress has been made in the clinical application of lung sound analysis techniques.

Objective: Using a new software program to analyze pediatric lung sounds using machine learning (ML), we conducted a lung sound survey study of 139 healthy 3-year-old children.

Subjects and methods: All cases were surveyed using the ATS-DLD questionnaire, which mainly included items related to a history of wheezing, diagnosis of asthma, and history of respiratory syncytial virus (RSV) infection, allergies and environment. The characteristics of the lung sounds were examined, along with the results of the questionnaire and lung sound parameters.

Results: Children with a history of wheezing showed a higher maximum inspiratory frequency (FAP₀), lower basal power (PAP₀) (p < 0.001 and p < 0.001, respectively), and lower RPF_50p and RPF_75p (p = 0.003 and p = 0.003, respectively), suggesting the enhancement of the high-pitched region of the lung sound spectrum. A similar tendency was observed in children with a history of asthma or RSV infection. Furthermore, in the group of children with a history of wheezing, those with a history of acute respiratory tract infection (ARI) within 1 week were found to have an enhancement of the high-pitched region relative to those without history of ARI.

Conclusions: By utilizing a new analysis software program using ML, we found that 3-year-old children with a history of wheezing or suspected asthma had characteristic lung sounds even when healthy.