Respiratory Disease Classification by CNN using MFCC

Abstract

Respiratory disease is a sort of sickness that produces a high death rate in rural or urban settings. Respiratory illnesses must be detected in advance and the rapid growth of deep learning over the last several years will lead the analysis and calculation of respiratory sound by computer computing power into a new trend of disease detection. Noting recent progress in the field of image classification, in which CNN's are utilized to categories high-precision pictures. A technique of classification of breathing sonority by CNN is proposed in this work, where it is trained. To this end, each audio sample was visually represented, enabling the identification of classification resources by applying the same methodologies to categories of high-precision pictures. We employed the Mel frequency cepstral coefficients method (MFCCs). We extracted resources with MFCC for every audio file in the dataset, meaning for every audio sample that we have an image representation. In the categorization of respiratory diseases utilized in the six classes accessible in the database, the approach described in this article achieved results over 93 percent. The six classifications are COPD (Chronic Pulmonary Obstructive Disease), Healthy, URTI (Upper Respiratory Tract Infection), Bronchiectasis, Pneumonia, Bronchiolitis.