Abstract
Background/Objectives: Respiratory diseases are common and result in high mortality, especially in the elderly, with pneumonia and chronic obstructive pulmonary disease (COPD). Auscultation of lung sounds using a stethoscope is a crucial method for diagnosis, but it may require specialized training and the involvement of pulmonologists. This study aims to assist medical professionals who are non-pulmonologist doctors in early screening for pneumonia and COPD by developing a smart stethoscope with cloud server-embedded machine learning to diagnose lung sounds. Methods: The smart stethoscope was developed using a Micro-Electro-Mechanical system (MEMS) microphone to record lung sounds in the mobile application and then send them wirelessly to a cloud server for real-time machine learning classification. Results: The model of the smart stethoscope classifies lung sounds into four categories: normal, pneumonia, COPD, and other respiratory diseases. It achieved an accuracy of 89%, a sensitivity of 89.75%, and a specificity of 95%. In addition, testing with healthy volunteers yielded an accuracy of 80% in distinguishing normal and diseased lungs. Moreover, the performance comparison between the smart stethoscope and two commercial auscultation stethoscopes showed comparable sound quality and loudness results. Conclusions: The smart stethoscope holds great promise for improving healthcare delivery in the post-COVID-19 era, offering the probability of the most likely respiratory conditions for early diagnosis of pneumonia, COPD, and other respiratory diseases. Its user-friendly design and machine learning capabilities provide a valuable resource for non-pulmonologist doctors by delivering timely, evidence-based diagnoses, aiding treatment decisions, and paving the way for more accessible respiratory care.