Considering Methods of Managing Noises Such as Coughs and Footsteps When Measuring Lung Sounds in Real Time

Abstract

This study aims to develop a system for automatically measuring lung sounds in asthma patients in real time without doctors, and for detecting unusual sounds occurring while patients are unconscious. When measuring lung sounds in real time, a considerable amount of noise from the body's internal environment can become an obstacle. Therefore, it is necessary to devise ways to manage the noise occurring during the measurement of lung sounds. In this study, we considered methods for addressing such noises, based on coughs and footsteps features (footsteps). The results demonstrated that the convolutional neural network (CNN) model could classify usual and unusual lung sounds even if the measured lung sounds contained footsteps, and that computers could correctly distinguish lung sounds from noises such as coughs and footsteps.