IEEE Access (Jan 2019)

Hemodynamic Quantitative Analysis Based on Doppler Ultrasound for Arteriovenous Shunt Stenosis Screening

  • Jian-Xing Wu,
  • Pi-Yun Chen,
  • Hsiao-Chuan Liu,
  • Chia-Hung Lin,
  • Shigao Chen,
  • K. Kirk Shung

DOI
https://doi.org/10.1109/ACCESS.2019.2955742
Journal volume & issue
Vol. 7
pp. 171765 – 171775

Abstract

Read online

Clinically, arteriovenous shunt (AVS) stenosis results in turbulent and pulsatile flow because of high resistance and pressure within a narrowed space inside a stenotic access. Palpation and ultrasound methods are primarily used (first-line examination) to rapidly screen the risk of the degree of stenosis (DOS). Therefore, quantitative hemodynamic analysis involving Doppler ultrasound is performed in patients suffering from AVS stenosis and undergoing long-term hemodialysis. Doppler ultrasound with a center frequency of 7.5 MHz can provide substantial resolution and sensitivity to the measurement of blood flow velocity within a range of depth of 20.0-30.0 mm and a scan diameter of 10.0 mm. A hemodynamic method is used to analyze blood flow through a hemodialysis access in terms of dimensionless numbers. In this study, velocities were measured using Doppler ultrasound at three specific sites in vessels, namely, arterial anastomosis, loop, and venous anastomosis sites. Dimensionless numbers, such as supracritical Reynolds numbers, critical peak Reynolds numbers, and resistive indices, are determined in accordance with parallel conditional expression-based rules to create decision trees for the rapid screening of the DOS at the abovementioned specific sites. For the enrolled subjects, results demonstrate that noninvasive hemodynamic analysis with Doppler ultrasound measurements and parallel decision trees has potential for the efficient screening of the DOS in patients suffering from AVS stenosis and undergoing long-term hemodialysis. Experimental results also indicate that the hit and true-positive rates of the proposed screening method in clinical indication are higher than those of the machine learning method.

Keywords