Respiratory sound analysis : performance of different system designs

• Main Author: Zhang, Yuxue
• Other Authors: Ser Wee
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: https://hdl.handle.net/10356/142780
• Institution: Nanyang Technological University

In many areas of the world, cardiopulmonary diseases are the main reason of leading of death. The respiration sounds are generated by air flow which are very valuable in the diagnose of cardiopulmonary diseases and respiratory diseases, because the sound would have some different sound features if there are diseases in respiratory or cardiovascular system. Therefore, the respiratory sound, or the pulmonary sound can be an important index on judging the cardiopulmonary diseases, and the system designs which could collect more valid and accurate data of the sound signal should be paid attention to. In this project, since different system designs would affect the sound signal collection which can further influence the judgement of the diseases, in order to improve the precision rate on the diagnose of the disease pulmonary edema, the performance evaluation of the different system designs is the emphasis and the purpose which would be finished by feature extraction, feature selection, classification and cross validation. To achieve the purpose of the project, initially managing and keeping the valid data to avoid some unstable factors, then preprocessing the remained data to simply the computation complexity of feature extraction. Using MFCC algorithm to extract 13 features to represent each sample, and then 3 more discriminate features were obtained to replacing 13 MFCC coefficients for each sample. Next, SVM algorithm was used on training the classifier model by using the selected features. Finally, evaluating the performance of the classifier models of 3 systems according to the indexes included sensitivity, specificity and accuracy which was calculated by 10-fold cross validation. In general, the conclusion of the performance of 3 systems was that system 2 was better than the other two systems with a highest accuracy 91.48% as well as the highest sensitivity and specificity 100%.