CNN-based Optimization for Lung Sound Classification with Mobile Accessibility

Integrating Artificial Intelligence (AI) into lung sound classification has markedly improved respiratory disease diagnosis by analyzing intricate patterns within audio data. This paper presents a two-phase approach for enhanced diagnosis of respiratory diseases. In the initial binary classification phase, lung sounds are identified as healthy or abnormal, enabling early exits for normal cases. The second phase categorizes abnormal sounds into one of nine specific diseases. The highest accuracy achieved for disease detection using binary classification and disease prediction using multi-class classification is 96.55% and 92.59%, respectively. The proposed model is developed as a user-friendly mobile application for seamless integration with the classification model. The mobile application allows users to upload lung sound files and predicts the top three potential diagnoses. The proposed two-phase strategy optimizes computational resources providing high accuracy and efficiency, while the mobile application extends the practical application of diagnostic models, with a system usability score above 73.21% showing good usability.