Classification of Arteriovenous Fistula Stenosis Using Shunt Murmur Analysis and Random Forest

Abstract

Although patients undergoing hemodialysis generally have shunts implanted within a body part, problems such as blood vessel stenosis can occur. Patients undergoing hemodialysis can conveniently check their own shunt function by listening to shunt murmurs. However, manually judging the shunt function is difficult and requires experience. In this study, we propose a method to classify shunt stenoses using Random Forest (RF). The resistance index (RI) obtained from the ultrasound system is used as a class label. The normalized cross-correlation coefficient, the ratio of frequency power to mel frequency cepstrum coefficient (MFCC), was used as a feature to train in an RF classifier. As a result, the classification accuracy of RI by RF was found to be higher than that achieved by a support vector machine (SVM).