High-accuracy classification of lung sounds using deep learning

Early diagnosis and treatment of respiratory system diseases are crucial for improving patients' quality of life and reducing mortality rates. Traditionally, doctors use stethoscopes to listen to patients' lung sounds and diagnose based on these sounds, which is a subjective and experience-dependent method. This study aims to classify lung sounds automatically. Audio files recorded with different filter types from patients with various lung diseases were used. The sounds were recorded from different points on the chest wall and diagnosed by specialists. Data augmentation techniques (pitch shift and time stretch) were employed to enhance model performance. Machine learning techniques were applied to extract MFCC features from these audio files, and a deep learning model with Bidirectional LSTM (BiLSTM) layers was used for classification. This study aims to develop a reliable and effective system for the early diagnosis of respiratory system diseases. The findings are intended to assist healthcare professionals in the diagnostic process and represent a significant advancement in patient care.