Lung Sound Based Pulmonary Disease Classification Using Deep Learning

Respiratory infections are disorders that damage the lungs and airways and make it difficult for patients to breathe. Any part of the respiratory system has the potential to become infected or sick, which could have a variety of negative effects. Early detection of lung disorders results in a person's successful treatment. Using electronically recorded lung sounds, pulmonary diseases can be diagnosed. MelFrequency Cepstral Coefficient (MFCC), one of the Librosa machine learning library features, is used to check that such a brief speech segment is sufficiently steady to permit effective modeling. In this research, Librosa is one of the better preprocessing techniques used when working with audio data. The CNN model is applied to the collected dataset and obtained a train accuracy of 95% and a test accuracy of 85% when trained on extracted MFCC features. And also used a number of performance evaluation indicators, including as F1score, accuracy, precision, and recall. The outcomes of this research enable the use of deep learning models in clinical settings to improve doctors' judgment when identifying lung disorders.