ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

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

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Abstract: FR-PO012

Assessment of Hemodialysis Arteriovenous Shunt (AV Shunt) Sounds by Using a Novel Electronic Stethoscope and Machine Learning Techniques

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science

Authors

  • Inagaki, Hiroko, Department of Medical Environment Innovation, Faculty of Medicine, University of Miyazaki,, Miyazaki, Japan
  • Tamura, Hiroki, Faculty of Engineering, University of Miyazaki, Miyazaki, Japan
  • Takeguchi, Hiromu, Graduate School of Engineering, University of Miyazaki, Miyazaki, Miyazaki, Japan
  • Endou, Yutaka, Miyazaki Seikyo Hospital, Miyazaki, Miyazaki, Japan
  • Hiroyuki, Fukumoto, Miyazaki Association Clinical Engineering Technologists, Miyazaki, Japan
  • Tanaka, Yasuomi, Togo Medikit Co., Ltd., Hyuga, Japan
  • Kibune, Kazuya, Department of Clinical Engineering Section, Tokatsu Dialysis Hospital & Clinic, Matsudo, Japan
  • Kikuchi, Masao, Division of Nephrology and Dialysis, University of Miyazaki Hospital, Miyazaki, Japan
  • Fujimoto, Shouichi, Department of Medical Environment Innovation, Faculty of Medicine, University of Miyazaki,, Miyazaki, Japan
Background

Hemodialysis AV shunt management is performed by screening for auscultated shunt sounds. However, determining stenosis with a stethoscope requires skill, and it is difficult to make assessment objectively and quantitatively. We linked the echo findings of AV shunt to audio data obtained from a novel electronic stethoscope (Togo Medikit Co., Ltd., Japan) and used these data to train AI, and then developed software that estimated two labels with AI. In this study, the usefulness of this software in clinical settings was examined by comparing the rating of trained staffs and AI for auscultatory sounds.

Methods

The quality and intensity of the shunt sounds were evaluated in six levels each by trained staff just before and after dialysis by auscultation mainly just above the shunt creation site. At the same time, 20 seconds of shunt sound data were electromagnetically recorded from each patient by a novel auscultation device. The sound quality rating by the staffs was 0~3 as bad, 4-5 as good, while the electromagnetically recorded sound data was rated as normal or abnormal using the developed software. Match rate, sensitivity, and specificity by both evaluation methods were calculated.

Results

We compared the ratings of trained nurses and AI for 191 auscultatory sounds. The matching rate between the two was 66.0%. 83.3% of the subjects were calculated as abnormal by AI, when trained staff judged as “bad”. 94.1% of the subjects were judged “good” by trained staff, when AI calculated as normal. Contrary to expectations, the rate of AI judging as normal was higher after dialysis than before, even though the intensity of the shunt sound was weaker after dialysis. When echo examination showed shunt blood flow reduction, it was more likely to be judged as abnormal by AI compared to the judgment by the staff.

Conclusion

In this study, AI rating for AV shunt sounds was considered to be at an acceptable level for clinical use. However, there are differences between trained staff and AI results for “normal” results. This reason for this may be that the AI label is based on the echo. This evaluation method using a new device is expected to become more useful with additional AI learning.