Sensor system design for respiratory sound classification

• Main Author: Wong, Shi Qi
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/141365
• Institution: Nanyang Technological University

Congestive Heart Failure is a life-threatening illness. There is no cure for such condition until today and thus, the only solution is early prevention by detecting the symptoms. Pulmonary Edema is one of the major symptoms of CHF whereby fluid accumulates in the lungs. It is caused by weak heart muscles or faulty valves when the heart can no longer pump blood effectively around the body. When we breathe, our lungs and airways produce a variety of sounds. The sounds produced from a person suffering from respiratory diseases are unique such as wheezes and crackles. These sounds can be used by doctors for diagnosis of the patients if they have water in the lungs. There are various methods to detect water in the lungs such as CT scan and MRI. However, these equipment are bulky, expensive and requires patients to visit the hospital. It is inconvenient and patients cannot use the equipment often. Therefore, a home-based early intervene device was introduced to allow users to be able to diagnose their health condition themselves. The aim of this project is to design a sound sensor that is able to provide better quality signals for processing by using a 3D printer. The effect of different designs of stethoscope acoustic housing will be investigated and the best design that is comparable with the current stethoscope design will be finalised. Several housing designs of different sizes and dimensions are created which includes bell-shaped, conical-shaped, cylindrical-shaped and parabolic-shaped. Each housing design are then integrated into microphone and 3.5mm audio jack cable for testing under the same setup and environment. The sound signals are recorded using voice memo application in a smart phone and saved in the computer for further analysis. Machine learning algorithms are then designed to process the sound signals. During the process, feature extraction using Mel-Frequency Cepstral Coefficient (MFCC), feature selection using fisher’s ratio (FR) and classification using support vector machine (SVM) are programmed into the MATLAB application. The final evaluation are through analysing their various SVM plot and their accuracy from SVM training. From the results, some of the housing shapes are comparable with the current stethoscope and will be further discussed in the later part of this report.