A Deep Learning Framework for Multiclass Categorization of Pulmonary Diseases

Abstract

Lung diseases such as pulmonary disease affect the lungs and respiratory organs which causes trouble in breathing and blocks the airflow in the body. This disease may be caused by infections, smoking tobacco, or other forms of air pollution. Pulmonary auscultation becomes the primary technique to identify the disease in the respiratory system. The sounds of air flowing inside and outside the lungs during the breathing process can be auscultated by a pulmonologist to identify the underlying disease in the respiratory system. With the immense development in technology, an automated framework for the multiclass categorization of respiratory disease is developed. The automated system uses Deep Learning techniques to do the classification. A Deep Neural Network such as Convolutional Neural Network is used for the classification of pulmonary disease using lung sounds. This will also aid in efficiently identifying diseases in a safe, non-invasive, environment-friendly, and sustainable way, improving the lives of the patients. This paper shows the implementation of Binary classification and Multiclass classification (8 classes) with the highest accuracy of 80% and 67% respectively. The obtained results are best to our knowledge when compared to the existing ones.