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: TH-PO027

Prediction of the Presence of Vesicoureteral Reflux Using Kidney Ultrasound Images with Deep Learning

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science

Authors

  • Morimoto, Yuichi, Kinki Daigaku, Higashiosaka, Osaka, Japan
  • Sugimoto, Keisuke, Kinki Daigaku, Higashiosaka, Osaka, Japan
  • Miyazaki, Kohei, Kinki Daigaku, Higashiosaka, Osaka, Japan
Background

In previous studies, detection of vesicoureteral reflux (VUR) and renal scarring by ultrasonography has been attempted, but the results have not yet been satisfactory.

Methods

Patients who underwent ultrasonography and VCUG examinations in the course of a thorough examination for congenital renal urinary tract abnormalities at our hospital from January 1, 2010 to January 1, 2021 were included in the study. We used a deep learning image classification method to create a discriminant model of VUR using kidney echo images as training data and VCUG examination results as teacher data. After creating the model using the training and validation data, the sensitivity and specificity were evaluated using the test data. The model was then evaluated for sensitivity and specificity using the test data. The model was visualized using the Grad-cam to determine the explanatory power of the model.

Results

In this study, 46 VCUG cases (male cases [57%], median age 4.5 years [0 months to 10 years]) were included. 23 of the 46 cases had VUR, 12 were Grade I-III and 11 were Grade IV-V. 230 echo images (115 each) from the groups with and without VUR were used for transition learning For VUR detection, the deep learning model had a sensitivity of 58%, specificity of 87.5%, positive predictive value of 82.4%, negative predictive value of 67.7%, and accuracy of 71.7% (Figure 1).

Conclusion

It was suggested that the presence of VUR could be predicted from renal echographic images and may contribute to the decision of VCUG indication. In addition, it has been reported that thickening of the renal pelvis wall due to reflux has been observed, and in the visualization of features performed with the model created in this study (Figure 1), similar areas may have contributed to the AI's judgment. If the presence and severity of VUR can be determined with a certain degree of accuracy from renal echo images, the pain and radiation exposure caused by VCUG may be avoided.

Features visualized by Gradcam