Mobile application-based pulmonary disease prediction using respiratory sound and deep learning

Pulmonary diseases are contagious illnesses that disrupt the respiratory system, often affecting the lungs. Diagnosing these diseases can be challenging due to similarities with other lung conditions. While many studies use pulmonary sounds for prediction, this study integrates patient medical history and respiratory sounds to enhance prediction accuracy using deep learning. By combining these data sources, a pulmonary disease prediction model was developed and integrated into a mobile app using TensorFlow Lite, improving accessibility. The model, leveraging Mel-spectrogram characteristics, achieved 97.0% accuracy, significantly higher than the 73.15% accuracy with only Spec-Augmentation. Evaluation by 10 experts on 30 use cases showed 26 accurate classifications, demonstrating the model's effectiveness and the benefits of using combined data for pulmonary disease prediction. In general, this study demonstrated that building a model using patient symptoms and pulmonary sound and embedding it in a mobile application improves the prediction of pulmonary disease significantly.