Quality Assessment of Respiratory Sounds Extracted from Self-Assembled Digital Stethoscopes

Abstract

Accurate assessment of respiratory sounds can aid the early detection of respiratory disorders such as crackles, wheezes, chronic obstructive pulmonary disease (COPD), and pneumonia. Typically, a stethoscope is used as a first aid to listen to respiratory sounds and initial diagnosis of underlying diseases. Unlike a traditional stethoscope, a digital stethoscope can offer recording of respiratory sounds and automatically diagnose abnormalities through machine learning technology. However, accurate machine learning models rely on good-quality data and features. The medical quality stethoscopes may provide high-quality data, however, are highly expensive. Alternatively, there are immense challenges in obtaining quality data from low-cost stethoscopes. The current work developed three inexpensive digital stethoscopes and compared the performance concerning six time and frequency domain features. The quality of extracted features was examined by Pearson's linear correlation coefficients. The results suggested that one of the low-cost stethoscopes exhibited 84% (5 out 6 features) highly correlated features. Based on the findings of this work, it may potentially help the researcher to carefully select low-cost stethoscopes for acquiring data.