Utilizing Neural Networks To Analyze Respiratory Sounds For The Detection Of Lung Diseases

Respiratory disorders, such as bronchitis and asthma, present significant global health challenges that require timely and accurate diagnosis for effective treatment. Traditional methods of diagnosing these conditions rely on manual analysis of lung sound recordings, which is time-consuming and prone to errors. This project introduces a novel automated diagnostic system leveraging advanced machine learning techniques to enhance diagnostic accuracy and efficiency. By employing deep learning methods, specifically Mel-Frequency Cepstral Coefficients (MFCCs), Continuous Short-Time Fourier Transform (CSTFT), and mel spectrograms for feature extraction, we developed a proprietary convolutional neural network (CNN) architecture. This architecture processes each audio feature type separately before integrating them into a cohesive model. Our extensive testing demonstrates that the system achieves a remarkable classification accuracy of 90.49% for respiratory illnesses. This innovative approach not only streamlines the diagnostic process but also provides a reliable tool for medical professionals, facilitating early detection and improved treatment planning for respiratory disorders.