Determination of pneumonia symptoms through acoustic analysis of cough sound and machine learning

Abstract

Cough is the most representative signals of the sound and vibration generated by the human body. The importance in smart healthcare is being emphasized due to the convenience of acquiring signals by non-invasive methods without visiting hospital. It also contains significant medical information related to the health status of respiratory system. In this study, various types of single cough sound were collected from adult patients with major respiratory diseases corresponding to pneumonia, acute bronchitis and chronic sinusitis. After dividing the collected data into two groups, pneumonia and non-pneumonia, the change aspects in sound pressure level and energy distribution for each frequency band were compared. Through this result, loudness and energy ratio are available as the objective diagnostic indicators for determining which group includes the respiratory disease. Therefore, these two characteristic factors were used as the input feature of machine learning algorithm with applying the data augmentation process for constructing big data set. By applying the algorithm to classification of data not used for training, it was found that the determination of pneumonia and non-pneumonia symptoms using cough sound could be performed with high accuracy.