ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: PO1033

Predicting Arteriovenous Graft Failure with Sound Signatures in Patients on Hemodialysis

Session Information

Category: Dialysis

  • 703 Dialysis: Vascular Access

Authors

  • Shien, Tiffany Wan-Hua, Kaohsiung Municipal United Hospital, Kaohsiung, Taiwan
  • Xu, Juan-Wei, National Kaohsiung Normal University - Yanchao Campus, Kaohsiung, Taiwan
  • Yeh, Yi-Ren, National Kaohsiung Normal University - Yanchao Campus, Kaohsiung, Taiwan
  • Lin, Hugo You-Hsien, Kaohsiung Municipal United Hospital, Kaohsiung, Taiwan
Background

The fast-growing prevalence of end stage renal disease leads to an increasing burden of population requiring dialysis worldwide. Specifically, patients on hemodialysis face the problem of maintaining their vascular accesses. Unfortunately, occurrence of stenosis and clots is not uncommon, especially in arteriovenous grafts (AVG). Graft longevity can be improved by effectively detecting and preventing these circumstances. The aim of this study is to develop a portable recording device that detects stenosis by extracting information from blood flow sounds.

Methods

Blood flow sounds were collected at four different locations on the arm, including venous and arterial ends of arteriovenous access. Measurements were conducted weekly, with four one-minute recordings per patient. A logistic regression model is used to analyze sound data. Recordings obtained prior to percutaneous transluminal angioplasty (PTA) procedures were labeled abnormal and those after PTA were labeled normal. Extracted features from each labeled recording include energy, spectrum, mel-frequency cepstrum, and chroma, as shown in Figure 1.

Results

In total, we have 109 labels, 25 of which are abnormal cases. Note that each case contains 4 separate recordings. For evaluation purposes, we randomly chose 75% of the labels as training cases and used the rest as testing cases. Each random trial compares single-location detection models to one integrated model, which combines data from all four locations. The trial was repeated 100 times. Our results in Table 1 indicate that arterial sounds are more informative than venous sounds in detecting stenosis. Note that the integrated model also significantly outperforms the other single-location models.

Conclusion

Our proposed model shows excellent performance in screening for AVG failure. This algorithm has potential to provide reliable and reproducible detection of vascular access abnormalities, optimizing AVG outcome and management for clinicians and patients.

Type I and Type II error rate of detection models
ModelVenous endArterial endGraft (v)Graft (a)Integrated model
Type I error rate0.5160.2790.5130.2980.124
Type II error rate0.2780.1180.2670.1610.060

Figure 1: Spectrum and roll-off frequency of (a) abnormal and (b) normal cases

Funding

  • Clinical Revenue Support