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-OR85

Development and Validation of a Deep-Learning Algorithm to Estimate Human Podometrics

Session Information

Category: Pathology and Lab Medicine

  • 1800 Pathology and Lab Medicine

Authors

  • Haruhara, Kotaro, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
  • Kawai, Hiroki, LPIXEL Inc., Tokyo, Japan
  • Kubo, Eisuke, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
  • Sasaki, Takaya, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
  • Okabayashi, Yusuke, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
  • Tsuboi, Nobuo, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
  • Yokoo, Takashi, Division of Nephrology and Hypertension, Department of Internal Medicine, The Jikei University School of Medicine, Tokyo, Japan
Background

Podocyte depletion is a common pathway in most progressive kidney diseases. We have recently established a method for estimating podometrics, namely podocyte number and size indices, using model-based stereology by measuring the number and diameter of podocyte nuclei on a double immunofluorescence for podocyte-specific markers. However, this method is time-consuming and lacks objectivity.

Methods

In the present study, we have developed a deep learning-based algorithm using modified UNet++ for segmenting podocyte nuclei and glomerulus areas in immunofluorescence images, which includes seed prediction for watershed analysis to enhance the identification of nuclei. Data for >43,000 podocyte nuclei on >1800 glomeruli from donors and autopsy kidneys that were manually counted were used as a ground truth. Approximately 70% of these data were randomly allocated for training deep learning algorithms, and the remaining approximately 30% of data were used as a test group. As a validation cohort, independent >500 images of podocyte immunofluorescence from patients with obesity-related glomerulopathy were used (Figure).

Results

In the test group consisting of donors and autopsy kidneys, our algorithm showed good accuracy with a precision of 0.92 and recall of 0.95 for podocyte identification. The correlation coefficient for the diameter of podocyte nuclei was 0.80. In the validation cohort, similar results were obtained. While the manual method to identify podocyte nuclei required approximately 30 minutes per glomerulus, our algorithm could detect much faster at a few seconds per glomerulus.

Conclusion

We developed and validated a deep-learning program to estimate podometrics with good accuracy and high throughput. This technology enables us to perform further podometric studies with a large sample size of subjects and glomeruli.

Funding

  • Government Support – Non-U.S.