Non-Invasive Tuberculosis Screening Based on Electronic Nose and Machine Learning

• Main Authors: KUO,YUAN-CHENG, 郭騵珹
• Other Authors: JONG,GWO-JIA
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/cpp8ba

碩士 === 國立高雄科技大學 === 電子工程系 === 107 === Tuberculosis (TB) is an infectious disease caused by mycobacterium tuberculosis. TB suspect (TBS) is someone who has TB symptoms. Such as cough for more than two weeks, accompanied by respiratory symptoms (shortness of breath, chest pain, hemoptysis) and other symptoms such as loss of appetite, weight loss, night sweats, and fatigue. TB suspect will produce volatile organic compound (VOCs) in their lung and come out when they get caught through an exhaled breath. In the medical field, TB diagnosis can be established on clinical symptoms, bacteriological examination, radiology, and other investigations. An electronic nose (e-Nose) is a device which has the function to mimic human nose. In biomedical field, e-Nose is used to detect and classify the chemical contain from human such as urine, etc. Related to the VOC in exhaled breath, e-Nose was used as non¬-invasive device to detect diseases based on those chemicals formed. In this thesis, we proposed an e-Nose to classify an exhaled breath of TBS and UTB. In general, e-Nose is formed by sensor array, hardware and artificial neural network (ANN). A sensor array was formed by Metal Oxide Semiconductor (MOS) sensors. Feature extraction was obtained from mean value of each sensor. The UTB and TBS were then processed in two methods. The first method was signal processing which has stages mean value as dataset and multilayer perceptron. The second method was image processing which has stages mean value as dataset converted in polar chart image. It is trained and classified using LeNet Network. The artificial neural network prediction was 95 % and LeNet was 85%.