A Study on Stenosis Detection Based on Non-contact Thrill Wave Imaging and Gradient-Boosting Decision Tree

Abstract

Hemodialysis therapy generally requires a special blood vessel called an arteriovenous fistula (AVF), which is surgically anastomosed between an artery and a vein. Since an AVF often becomes stenosis, palpation is used to palpate the vessel wall vibrations, which is called thrill wave, before and after hemodialysis treatment. This method is widely used, especially in Japan, because of its simplicity. However, several problems in the palpation has been pointed out in terms of reliability because the palpation requires contact diagnosis. In order to solve the problems in the conventional contact palpation, we developed a thrill wave measurement device using non-contact imaging based on an optical technology. Then, we introduced a gradient-boosting decision tree algorithm to detect stenosis in AVFs. The experimental results showed that true positive rate (TPR) = 92.3%, true negative rate (TNR) = 76.7%, false positive rate (FPR) = 7.7% and false negative rate (FNR) = 23.3% to identify normal and stenotic AVFs.