Application of Empirical Mode Decomposition and Machine Learning to Arteriovenous Graft Occlusion Analysis for Hemodialysis Patients

• Main Authors: Yu-YaoWang, 王昱堯
• Other Authors: Kuo-Sheng Cheng
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/g6c77m

碩士 === 國立成功大學 === 生物醫學工程學系 === 106 === The AV access is usually evaluated by feeling thrill and pulsation through palpation, listening for the bruit by using a stethoscope, Doppler ultrasound imaging, or angiography, etc. However, these techniques require specific equipment and operator. Phonoangiography is a noninvasive tool for identifying vascular diameter change. In this study, a mock model has been set up to simplify the simulation of blood flow condition. Phonographic signal is recorded by electronic stethoscope and further signal processed. The relationship of phonographic signals and stenotic lesions is studied. Early detection of hemodialysis access problems such as stenosis and thrombosis is very important issue. The purpose of this study is to develop a phonographic system to evaluate arteriovenous shunt (AVS) stenosis of hemodialysis patients. The degree of stenosis (DOS) is used as an index to classify the AV access condition, and is determined by the narrowing percentage of normal vessels. The empirical mode decomposition (EMD) method is applied to analyze the relationship between DOS and spectrogram. After feature extraction, use machine learning to train prediction model and classify. Verification is based on Doppler ultrasound which is the golden standard in clinical application. In 22 cases, KNN and SVM show 90.9% and 85.7% accuracy respectively, it proved that empirical mode decomposition is feasible in feature extraction. This noninvasive method may be useful and potential for early detection in home-care use.