Respiratory sound classification : effect of heart sound

Abstract

Singapore has known to be a country with rising amount of population and coupled with a backdrop of ageing population. This proved to be a stress and puts a toll on the medical infrastructure. In a recent study conducted by the Ministry of health, it was revealed that the 2nd most prominent cause for death in Singapore is due to Pneumonia, just behind cancer. [1] Infants and young children, people over the age of 65, and people with chronic disease are at greatest risk of developing severe pneumonia which can be life-threatening. [2]

Although there might be other methods such as X-ray and CT scan to aid in the detection of any presence of water in the lungs, these can be pretty big and costly, which implies that it might not be easily available to everyone. Hence, this presents the need for research in automatic detection approach to anaylazing various lung sounds and attempt to classify them on the distinctive features of theirs. This would then help to better improve accuracy of the detection for these lung related diseases.

With that in mind, the project tried to study and analyse the performance of involvement of signal processing together with machine learning. The data collected will be audio samples collected from patients in Tan Tock Seng Hospital with the aid of professionals. Features extraction, features selection, and model classification will then be utilized. As raw data in audio format cannot be understand by models directly, this required the conversion into representable format of music, and thus audio is passed through Mel-Frequency Cepstrum Coefficients to enable features extraction. The extracted features are then manually selected in a way in which will contribute the most to prediction output. This will then be passed into Support Vector Machine to aid in analyzing the selected features. It will then be evaluated based on accuracy, sensitivity and specificity.