Abstract:

At present, a large number of people lose their lives due to different respiratory diseases every day. Respiratory Sound Analysis has been a key tool to accurately detect these types of diseases. Earlier, manual detection of respiratory sounds was used but it is not feasible to detect various lung diseases due to various reasons like audio quality and perceptions of different doctors. Modern computer aided analysis helps to identify the diseases better with the sound and earlier treatment can be given to patients. These respiratory sound diseases include Asthma, Bronchitis, Pneumonia, COPD and URTI. The prediction with decision trees gives an accuracy rate of 88 percent, support vector machine gives an accuracy rate of the 82 percent and logistic regression gives an accuracy of 72 percent. A CNN model built and trained using the spectrogram images of audio files gave an accuracy rate of the 82 percent. Thus the proposed system detects the disease more earlier than manual detection with the help of respiratory sounds and tells us the exact lung disease in a better way.

Keywords: