Lung Sound Classification for Respiratory Disease

Abstract

Respiratory illnesses are still among the biggest health problems around the world, and catching them early is key to avoiding serious issues. Listening to lung sounds is a widely used method for identifying respiratory diseases. However, manual examination depends heavily on a clinician’s experience, may differ from one doctor to another, and can be unreliable in noisy surroundings. To overcome these challenges, this work presents an automated lung sound classification system based on deep learning techniques. Important acoustic features are extracted using Mel-Frequency Cepstral Coefficients (MFCCs), which are then analyzed using a hybrid CNN-LSTM model to categorize sounds into six classes: COPD, Pneumonia, Bronchiectasis, Bronchiolitis, Upper Respiratory Tract Infection (URTI), and Normal. The model was trained, tested, and optimized to ensure reliable performance across varied conditions. Additionally, Grad-CAM is integrated to highlight the sound regions that influence the model’s predictions, improving transparency and interpretability. The proposed system offers accurate results with a user-friendly interface, making it suitable for real-time clinical use. Overall, this approach supports early detection of respiratory illnesses and has strong potential for deployment in regions with limited medical resources.